Weekly TMLR digest for Aug 17, 2025
TMLR
New certifications
==================
Featured Certification: Random Erasing vs. Model Inversion: A Promising Defense or a False Hope?
Viet-Hung Tran, Ngoc-Bao Nguyen, Son T. Mai, Hans Vandierendonck, Ira Assent, Alex Kot, Ngai-Man Cheung
https://openreview.net/forum?id=S9CwKnPHaO
---
Accepted papers
===============
Title: LumiNet: Perception-Driven Knowledge Distillation via Statistical Logit Calibration
Authors: Md. Ismail Hossain, M M Lutfe Elahi, Sameera Ramasinghe, Ali Cheraghian, Fuad Rahman, Nabeel Mohammed, Shafin Rahman
Abstract: In the knowledge distillation literature, feature-based methods have dominated due to their ability to effectively tap into extensive teacher models. In contrast, logit-based approaches, which aim to distill `dark knowledge' from teachers, typically exhibit inferior performance compared to feature-based methods. To bridge this gap, we present LumiNet, a novel knowledge distillation algorithm designed to enhance logit-based distillation. We introduce the concept of `perception', aiming to calibrate logits based on the model's representation capability. This concept addresses overconfidence issues in the logit-based distillation method while also introducing a novel method to distill knowledge from the teacher. It reconstructs the logits of a sample/instances by considering relationships with other samples in the batch. LumiNet excels on benchmarks like CIFAR-100, ImageNet, and MSCOCO, outperforming the leading feature-based methods, e.g., compared to KD with ResNet18 and MobileNetV2 on ImageNet, it shows improvements of 1.5\% and 2.05\%, respectively.
URL: https://openreview.net/forum?id=3rU1lp9w2l
---
Title: Task Diversity Shortens the In-Context Learning Plateau
Authors: Jaeyeon Kim, Sehyun Kwon, Joo Young Choi, Jongho Park, Jaewoong Cho, Jason D. Lee, Ernest K. Ryu
Abstract: In-context learning (ICL) describes a language model's ability to generate outputs based on a set of input demonstrations and a subsequent query. To understand this remarkable capability, researchers have studied simplified, stylized models. These studies have consistently observed long loss plateaus, during which models exhibit minimal improvement, followed by a sudden, rapid surge of learning. In this work, we reveal that training on multiple diverse ICL tasks simultaneously shortens the loss plateaus, making each task easier to learn. This finding is surprising as it contradicts the natural intuition that the combined complexity of multiple ICL tasks would lengthen the learning process, not shorten it. Our result suggests that the recent success in large-scale training of language models may be attributed not only to the richness of the data at scale but also to the easier optimization (training) induced by the diversity of natural language training data.
URL: https://openreview.net/forum?id=7t5DzaJOdB
---
Title: CAREL: Instruction-guided reinforcement learning with cross-modal auxiliary objectives
Authors: Armin Saghafian, Amirmohammad Izadi, Negin Hashemi Dijujin, Mahdieh Soleymani Baghshah
Abstract: Grounding the instruction in the environment is a key step in solving language-guided goal-reaching reinforcement learning problems. In automated reinforcement learning, a key concern is to enhance the model's ability to generalize across various tasks and environments. In goal-reaching scenarios, the agent must comprehend the different parts of the instructions within the environmental context in order to complete the overall task successfully. In this work, we propose \textbf{CAREL} (\textit{\textbf{C}ross-modal \textbf{A}uxiliary \textbf{RE}inforcement \textbf{L}earning}) as a new framework to solve this problem using auxiliary loss functions inspired by video-text retrieval literature and a novel method called instruction tracking, which automatically keeps track of progress in an environment. The results of our experiments suggest superior sample efficiency and systematic generalization for this framework in multi-modal reinforcement learning problems.
URL: https://openreview.net/forum?id=zJUEYr5X1X
---
Title: Unifying Self-Supervised Clustering and Energy-Based Models
Authors: Emanuele Sansone, Robin Manhaeve
Abstract: Self-supervised learning excels at learning representations from large amounts of data. At the same time, generative models offer the complementary property of learning information about the underlying data generation process. In this study, we aim at establishing a principled connection between these two paradigms and highlight the benefits of their complementarity. In particular, we perform an analysis of self-supervised learning objectives, elucidating the underlying probabilistic graphical models and presenting a standardized methodology for their derivation from first principles. The analysis suggests a natural means of integrating self-supervised learning with likelihood-based generative models. We instantiate this concept within the realm of cluster-based self-supervised learning and energy models, introducing a lower bound proven to reliably penalize the most important failure modes and unlocking full unification. Our theoretical findings are substantiated through experiments on synthetic and real-world data, including SVHN, CIFAR10, and CIFAR100, demonstrating that our objective function allows to jointly train a backbone network in a discriminative and generative fashion, consequently outperforming existing self-supervised learning strategies in terms of clustering, generation and out-of-distribution detection performance by a wide margin. We also demonstrate that the solution can be integrated into a neuro-symbolic framework to tackle a simple yet non-trivial instantiation of the symbol grounding problem.
URL: https://openreview.net/forum?id=NW0uKe6IZa
---
Title: MESSI: A Multi-Elevation Semantic Segmentation Image Dataset of an Urban Environment
Authors: Barak Pinkovich, Boaz Matalon, Ehud Rivlin, Hector Rotstein
Abstract: This paper presents a Multi-Elevation Semantic Segmentation Image (MESSI) dataset. A reduced version of the dataset has been published at https://github.com/messi-dataset/ for reviewing purposes (due to the anonymity requirement). The full dataset will be made available at the time of the decision. MESSI comprises 2525 images taken by a drone flying over dense urban environments. MESSI is unique in two main features. First, it contains images from various altitudes (both with horizontal and vertical trajectories), allowing us to investigate the effect of depth on semantic segmentation. Second, it includes images taken from several different urban regions (at different altitudes). This is important since the variety covers the visual richness captured by a drone's 3D flight, performing horizontal and vertical maneuvers. MESSI contains images annotated with location, orientation, and the camera's intrinsic parameters. It can be used to train a deep neural network for semantic segmentation or other applications of interest. This paper describes the dataset and provides annotation details. It also explains how semantic segmentation was performed using several neural network models and shows several relevant statistics. MESSI will be published in the public domain to serve as an evaluation benchmark for semantic segmentation using images captured by a drone or similar vehicle flying over a dense urban environment.
URL: https://openreview.net/forum?id=ayWqZ1wyIv
---
Title: Combining Machine Learning Defenses without Conflicts
Authors: Vasisht Duddu, Rui Zhang, N. Asokan
Abstract: Machine learning (ML) models require protection against various risks to security, privacy, and fairness. Real-life ML models need simultaneous protection against multiple risks, necessitating combining multiple defenses effectively, without incurring significant drop in the effectiveness of the constituent defenses. We present a systematization of existing work based on how defenses are combined, and how they interact. We then identify unexplored combinations, and evaluate combination techniques to identify their limitations. Using these insights, we present, Def\Con, a combination technique which is (a) accurate (correctly identifies whether a combination is effective or not), (b) scalable (allows combining multiple defenses), (c) non-invasive (allows combining existing defenses without modification), and (d) general (is applicable to different types of defenses). We show that Def\Con achieves 90% accuracy on eight combinations from prior work, and 86% in 30 unexplored combinations evaluated empirically.
URL: https://openreview.net/forum?id=C7FgsjfFRC
---
Title: How does overparametrization affect performance on minority groups?
Authors: Saptarshi Roy, Subha Maity, Songkai Xue, Mikhail Yurochkin, Yuekai Sun
Abstract: The benefits of overparameterization for the overall performance of modern machine learning (ML) models are well known. However, the effect of overparameterization at a more granular level of data subgroups is less understood. Recent empirical studies demonstrate encouraging results: (i) when groups are not known, overparameterized models trained with empirical risk minimization (ERM) perform better on minority groups; (ii) when groups are known, ERM on data subsampled to equalize group sizes yields state-of-the-art worst-group accuracy in the overparameterized regime. In this paper, we complement these empirical studies with a theoretical investigation of the risk of overparameterized random feature regression models on minority groups with identical feature distribution as the majority group. In a setting in which the regression functions for the majority and minority groups are different, we show that overparameterization either improves or does not harm the asymptotic minority group performance under the ERM setting when the features are distributed uniformly over the sphere.
URL: https://openreview.net/forum?id=POunezXgvF
---
Title: Beyond Grids: Multi-objective Bayesian Optimization With Adaptive Discretization
Authors: Andi Nika, Sepehr Elahi, Cagin Ararat, Cem Tekin
Abstract: We consider the problem of optimizing a vector-valued objective function $\boldsymbol{f}$ sampled from a Gaussian Process (GP) whose index set is a well-behaved, compact metric space $(\mathcal{X},d)$ of designs. We assume that $\boldsymbol{f}$ is not known beforehand and that evaluating $\boldsymbol{f}$ at design $x$ results in a noisy observation of $\boldsymbol{f}(x)$. Since identifying the Pareto optimal designs via exhaustive search is infeasible when the cardinality of $\mathcal{X}$ is large, we propose an algorithm, called Adaptive $\boldsymbol{\epsilon}$-PAL, that exploits the smoothness of the GP-sampled function and the structure of $(\mathcal{X},d)$ to learn fast. In essence, Adaptive $\boldsymbol{\epsilon}$-PAL employs a tree-based adaptive discretization technique to identify an $\boldsymbol{\epsilon}$-accurate Pareto set of designs in as few evaluations as possible. We provide both information-type and metric dimension-type bounds on the sample complexity of $\boldsymbol{\epsilon}$-accurate Pareto set identification. We also experimentally show that our algorithm outperforms other Pareto set identification methods.
URL: https://openreview.net/forum?id=Wq150HaRVE
---
Title: Controlling Statistical, Discretization, and Truncation Errors in Learning Fourier Linear Operators
Authors: Unique Subedi, Ambuj Tewari
Abstract: We study learning-theoretic foundations of operator learning, using the linear layer of the Fourier Neural Operator architecture as a model problem. First, we identify three main errors that occur during the learning process: statistical error due to finite sample size, truncation error from finite rank approximation of the operator, and discretization error from handling functional data on a finite grid of domain points. Finally, we analyze a Discrete Fourier Transform (DFT) based least squares estimator, establishing both upper and lower bounds on the aforementioned errors.
URL: https://openreview.net/forum?id=A2sHNGcjLO
---
Title: Table Foundation Models: on knowledge pre-training for tabular learning
Authors: Myung Jun Kim, Félix Lefebvre, Gaëtan Brison, Alexandre Perez-Lebel, Gaël Varoquaux
Abstract: Table foundation models bring high hopes to data science: pre-trained on tabular data to embark knowledge or priors, they should facilitate downstream tasks on tables. One specific challenge is that of data semantics: numerical entries take their meaning from context, *e.g.*, column name. The traditional approach combines column-specific data preparation with tree-based models that adapt to column specificities. Pre-trained neural networks that jointly model column names and table entries have recently boosted prediction accuracy. While these models outline the promises of world knowledge to interpret table values, they lack the convenience of popular foundation models in text or vision. Indeed, they must be fine-tuned to bring benefits, come with sizeable computation costs, and cannot easily be reused or combined with other architectures. Here we introduce TARTE, a foundation model that transforms tables to knowledge-enhanced vector representations using the string to capture semantics. Pre-trained on large relational data, TARTE yields representations that facilitate subsequent learning with little additional cost. These representations can be fine-tuned or combined with other learners, giving models that push the state-of-the-art prediction performance and improve the prediction/computation performance trade-off. Specialized to a task or a domain, TARTE gives domain-specific representations that facilitate further learning. Our study demonstrates an effective approach to knowledge pre-training for tabular learning.
URL: https://openreview.net/forum?id=QV4P8Csw17
---
Title: ReFeR: Improving Evaluation and Reasoning through Hierarchy of Models
Authors: Yaswanth Narsupalli, Abhranil Chandra, Sreevatsa Muppirala, Manish Gupta, Pawan Goyal
Abstract: Assessing the quality of generative model outputs from large language models (LLMs) or vision-language models (VLMs), poses significant challenges. Traditional evaluation methods either rely on human assessment which is resource-intensive and not scalable or on automatic metrics that often correlate poorly with human preferences. Another approach is to train dedicated neural evaluators, but this typically requires substantial training data and compute. In this study, we thus introduce ReFeR, a tuning-free framework for evaluating generative outputs including both text and images, using a two-level hierarchy of pre-trained LLM and VLM evaluators. This multi-agent hierarchical strategy leverages additional compute at inference time by orchestrating multiple models and utilizing the increased test-time reasoning to boost performance. By having models themselves provide feedback and final judgments, ReFeR reduces the dependence on human evaluation. We rigorously evaluate ReFeR on four diverse evaluation benchmarks, where it surpasses prior methods in accuracy while also generating constructive feedback useful for downstream distillation and self-improvement via finetuning. Interestingly, ReFeR is also applicable for reasoning tasks - experiments on four reasoning benchmarks show ReFeR’s superior collective reasoning abilities. We present two variants of the framework: ReFeR-Turbo, optimized for accelerated performance, and ReFeR-Lite, offering a more test-time compute efficient solution. ReFeR-Lite is $\sim12-14\times$ more compute efficient than previous works while being comparably accurate to ReFeR-Turbo.
URL: https://openreview.net/forum?id=otSHFe8wTf
---
Title: A stochastic gradient descent algorithm with random search directions
Authors: Eméric Gbaguidi
Abstract: Stochastic coordinate descent algorithms are efficient methods in which each iterate is obtained by fixing most coordinates at their values from the current iteration, and approximately minimizing the objective with respect to the remaining coordinates. However, this approach is usually restricted to canonical basis vectors of $\mathbb{R}^d$. In this paper, we develop the class of stochastic gradient descent algorithms with random search directions. These methods use the directional derivative of the gradient estimate following more general random vectors. We establish the almost sure convergence of these algorithms with decreasing step. We further investigate their central limit theorem and pay particular attention to analyze the impact of the search distributions on the asymptotic covariance matrix. We also provide non-asymptotic $\mathbb{L}^p$ rates of convergence.
URL: https://openreview.net/forum?id=npER8AaLSb
---
Title: FoldDiff: Folding in Point Cloud Diffusion
Authors: Yuzhou Zhao, Juan Matias Di Martino, Amirhossein Farzam, Guillermo Sapiro
Abstract: Diffusion denoising has emerged as a powerful approach for modeling data distributions, treating data as particles with their position and velocity modeled by a stochastic diffusion process. While this framework assumes data resides in a fixed vector spaces (e.g., images as pixel-ordered vectors), point clouds present unique challenges due to their unordered representation. Existing point cloud diffusion methods often rely on voxelization to address this issue, but this approach is computationally expensive, with cubically scaling complexity. In this work, we investigate the misalignment between point cloud irregularity and diffusion models, analyzing it through the lens of denoising implicit priors. First, we demonstrate how the unknown permutations inherent in point cloud structures disrupt denoising implicit priors. To address this, we then propose a novel folding-based approach that reorders point clouds into a permutation-invariant grid, enabling diffusion to be performed directly on the structured representation. This construction is exploited both globally and locally. Globally, \reviewcdmS{folded objects can represent point cloud objects} in a fixed vector space (like images), therefore it enables us to extend the work of denoising as implicit priors to point clouds. \reviewcdmS{Locally, the folded tokens are} efficient and novel token representations that can improve existing transformer-based point cloud diffusion models. Our experiments show that the proposed folding operation integrates effectively with both denoising implicit priors as well as advanced diffusion architectures, such as UNet and Diffusion Transformers (DiTs). Notably, DiT with \reviewcdmS{locally} folded tokens achieves competitive generative performance compared to state-of-the-art models while significantly reducing training and inference costs relative to voxelization-based methods.
URL: https://openreview.net/forum?id=pmRabMH1JW
---
Title: Personalized Federated Learning via Low-Rank Matrix Optimization
Authors: Ali Dadras, Sebastian U Stich, Alp Yurtsever
Abstract: Personalized Federated Learning (pFL) has gained significant attention for building a suite of models tailored to different clients. In pFL, the challenge lies in balancing the reliance on local datasets, which may lack representativeness, against the diversity of other clients' models, whose quality and relevance are uncertain. Focusing on the clustered FL scenario, where devices are grouped based on similarities in their data distributions without prior knowledge of cluster memberships, we develop a mathematical model for pFL using low-rank matrix optimization. Building on this formulation, we propose a pFL approach leveraging the Burer-Monteiro factorization technique. We examine the convergence guarantees of the proposed method and present numerical experiments on training deep neural networks, demonstrating the empirical performance of the proposed method in scenarios where personalization is crucial.
URL: https://openreview.net/forum?id=DFJu1QB2Nr
---
Title: Node-Level Data Valuation on Graphs
Authors: Simone Antonelli, Aleksandar Bojchevski
Abstract: How much is a node worth? We answer this question using an emerging set of data valuation techniques, where the value of a data point is measured via its marginal contribution when added to the (training) dataset. Data valuation has been primarily studied in the i.i.d. setting, giving rise to methods like influence functions, leave-one-out estimation, data Shapley, and data Banzhaf. We conduct a comprehensive study of data valuation approaches applied to graph-structured models such as graph neural networks in a semi-supervised transductive setting. Since all nodes (labeled and unlabeled) influence both training and inference we construct various scenarios to understand the diverse mechanisms by which nodes can impact learning. We show that the resulting node values can be used to identify (positively and negatively) influential nodes, quantify model brittleness, detect poisoned data, and accurately predict counterfactuals.
URL: https://openreview.net/forum?id=tNyApIqDSJ
---
Title: Unified Wisdom: Harnessing Collaborative Learning to Improve Efficacy of Knowledge Distillation
Authors: Atharva Abhijit Tambat, Durga S, Ganesh Ramakrishnan, Pradeep Shenoy
Abstract: Knowledge distillation (KD), which involves training a smaller student model to approximate the predictions of a larger teacher model is useful in striking a balance between model accuracy and computational constraints. However, KD has been found to be ineffective when the teacher and student models have a significant capacity gap. In this work, we address this issue via “meta-collaborative distillation” (MC-Distil), where students of varying capacities collaborate during distillation. Using a “coordinator” network (C-Net), MC-Distil enables mutual learning among students as a meta-learning task. Our insight is that C-Net learns from each student’s performance and training instance characteristics, allowing students of different capacities to improve together. Our method enhances student accuracy for all students, surpassing state-of-the-art baselines, including multi-step distillation, consensus enforcement, and teacher re-training. We achieve average gains of 2.5% on CIFAR-100 and 2% on Tiny ImageNet datasets, consistently across diverse student sizes, teacher sizes, and architectures. Notably, larger students benefiting through meta-collaboration with smaller students is a novel idea. MC-Distil excels in training superior student models under real-world conditions such as label noise and domain adaptation. Our approach also yields consistent improvements on the MS COCO object detection benchmark and introduces only a modest 5% computational overhead during training, with no additional cost at inference.
URL: https://openreview.net/forum?id=Zj9bb8aQNg
---
Title: On the Convergence of SVGD in KL divergence via Approximate gradient flow
Authors: Masahiro Fujisawa, Futoshi Futami
Abstract: This study investigates the convergence of Stein variational gradient descent (SVGD), which is used to approximate a target distribution based on a gradient flow on the space of probability distributions. The existing studies mainly focus on the convergence in the kernel Stein discrepancy, which doesn't imply weak convergence in many practical settings. To address this issue, we propose to introduce a novel analytical approach called $(\epsilon,\delta)$-approximate gradient flow, extending conventional concepts of approximation error for the Wasserstein gradient. With this approach, we show the sub-linear convergence of SVGD in Kullback--Leibler divergence under the discrete time and infinite particle settings. Finally, we validate our theoretical findings through several numerical experiments.
URL: https://openreview.net/forum?id=AG1zXt5aoA
---
Title: BELLA: Black-box model Explanations by Local Linear Approximations
Authors: Nedeljko Radulovic, Albert Bifet, Fabian M. Suchanek
Abstract: Understanding the decision-making process of black-box models has become not just a legal requirement, but also an additional way to assess their performance. However, the state of the art post-hoc explanation approaches for regression models rely on synthetic data generation, which introduces uncertainty and can hurt the reliability of the explanations. Furthermore, they tend to produce explanations that apply to only very few data points. In this paper, we present BELLA, a deterministic model-agnostic post-hoc approach for explaining the individual predictions of regression black-box models. BELLA provides explanations in the form of a linear model trained in the feature space. BELLA maximizes the size of the neighborhood to which the linear model applies so that the explanations are accurate, simple, general, and robust.
URL: https://openreview.net/forum?id=F9Kv96KcwM
---
Title: Emergent Symbol-like Number Variables in Artificial Neural Networks
Authors: Satchel Grant, Noah Goodman, James Lloyd McClelland
Abstract: What types of numeric representations emerge in neural systems, and what would a satisfying answer to this question look like? In this work, we interpret Neural Network (NN) solutions to sequence based number tasks through a variety of methods to understand how well we can interpret them through the lens of interpretable Symbolic Algorithms (SAs)—precise algorithms describable by rules operating on typed, mutable variables. We use GRUs, LSTMs, and Transformers trained using Next Token Prediction (NTP) on tasks where the correct tokens depend on numeric information only latent in the task structure. We show through multiple causal and theoretical methods that we can interpret raw NN activity through the lens of simplified SAs when we frame the neural activity in terms of neural subspaces rather than individual neurons. Using Distributed Alignment Search (DAS), we find that, depending on network architecture, dimensionality, and task specifications, alignments with SA’s can be very high, while other times they can be only approximate, or fail altogether. We extend our analytic toolkit to address the failure cases by expanding the DAS framework to a broader class of alignment functions that more flexibly capture NN activity in terms of interpretable variables from SAs, and we provide theoretic and empirical explorations of Linear Alignment Functions (LAFs) in contrast to the preexisting Orthogonal Alignment Functions (OAFs). Through analyses of specific cases we confirm the usefulness of causal interventions on neural subspaces for NN interpretability, and we show that recurrent models can develop graded, symbol-like number variables within their neural activity. We further show that shallow Transformers learn very different solutions than recurrent networks, and we prove that such models must use anti-Markovian solutions—solutions that do not rely on cumulative, Markovian hidden states—in the absence of sufficient attention layers.
URL: https://openreview.net/forum?id=YPnYpiru5W
---
Title: Loss Landscape Degeneracy and Stagewise Development in Transformers
Authors: Jesse Hoogland, George Wang, Matthew Farrugia-Roberts, Liam Carroll, Susan Wei, Daniel Murfet
Abstract: Deep learning involves navigating a high-dimensional loss landscape over the neural network parameter space. Over the course of training, complex computational structures form and re-form inside the neural network, leading to shifts in input/output behavior. It is a priority for the science of deep learning to uncover principles governing the development of neural network structure and behavior. Drawing on the framework of singular learning theory, we propose that model development is deeply linked to degeneracy in the local geometry of the loss landscape. We investigate this link by monitoring loss landscape degeneracy throughout training, as quantified by the local learning coefficient, for a transformer language model and an in-context linear regression transformer. We show that training can be divided into distinct periods of change in loss landscape degeneracy, and that these changes in degeneracy coincide with significant changes in the internal computational structure and the input/output behavior of the transformers. This finding provides suggestive evidence that degeneracy and development are linked in transformers, underscoring the potential of a degeneracy-based perspective for understanding modern deep learning.
URL: https://openreview.net/forum?id=45qJyBG8Oj
---
Title: Does equivariance matter at scale?
Authors: Johann Brehmer, Sönke Behrends, Pim De Haan, Taco Cohen
Abstract: Given large datasets and sufficient compute, is it beneficial to design neural architectures for the structure and symmetries of each problem? Or is it more efficient to learn them from data? We study empirically how equivariant and non-equivariant networks scale with compute and training samples. Focusing on a benchmark problem of rigid-body interactions and on general-purpose transformer architectures, we perform a series of experiments, varying the model size, training steps, and dataset size. We find evidence for three conclusions. First, equivariance improves data efficiency, but training non-equivariant models with data augmentation can close this gap given sufficient epochs. Second, scaling with compute follows a power law, with equivariant models outperforming non-equivariant ones at each tested compute budget. Finally, the optimal allocation of a compute budget onto model size and training duration differs between equivariant and non-equivariant models.
URL: https://openreview.net/forum?id=wilNute8Tn
---
Title: Continuous Parallel Relaxation for Finding Diverse Solutions in Combinatorial Optimization Problems
Authors: Yuma Ichikawa, Hiroaki Iwashita
Abstract: Finding the optimal solution is often the primary goal in combinatorial optimization (CO). However, real-world applications frequently require diverse solutions rather than a single optimum, particularly in two key scenarios. The first scenario occurs in real-world applications where strictly enforcing every constraint is neither necessary nor desirable. Allowing minor constraint violations can often lead to more cost-effective solutions. This is typically achieved by incorporating the constraints as penalty terms in the objective function, which requires careful tuning of penalty parameters. The second scenario involves cases where CO formulations tend to oversimplify complex real-world factors, such as domain knowledge, implicit trade-offs, or ethical considerations. To address these challenges, generating (i) penalty-diversified solutions by varying penalty intensities and (ii) variation-diversified solutions with distinct structural characteristics provides valuable insights, enabling practitioners to post-select the most suitable solution for their specific needs. However, efficiently discovering these diverse solutions is more challenging than finding a single optimal one. This study introduces Continual Parallel Relaxation Annealing (CPRA), a computationally efficient framework for unsupervised-learning (UL)-based CO solvers that generates diverse solutions within a single training run. CPRA leverages representation learning and parallelization to automatically discover shared representations, substantially accelerating the search for these diverse solutions. Numerical experiments demonstrate that CPRA outperforms existing UL-based solvers in generating these diverse solutions while significantly reducing computational costs.
URL: https://openreview.net/forum?id=ix33zd5zCw
---
Title: Are Domain Generalization Benchmarks with Accuracy on the Line Misspecified?
Authors: Olawale Elijah Salaudeen, Nicole Chiou, Shiny Weng, Sanmi Koyejo
Abstract: Spurious correlations, unstable statistical shortcuts a model can exploit, are expected to degrade performance out-of-distribution (OOD). However, across many popular OOD generalization benchmarks, vanilla empirical risk minimization (ERM) often achieves the highest OOD accuracy. Moreover, gains in in-distribution accuracy generally improve OOD accuracy, a phenomenon termed accuracy on the line, which contradicts the expected harm of spurious correlations. We show that these observations are an artifact of misspecified OOD datasets that do not include shifts in spurious correlations that harm OOD generalization, the setting they are meant to evaluate. Consequently, current practice evaluates "robustness" without truly stressing the spurious signals we seek to eliminate; our work pinpoints when that happens and how to fix it. Contributions. (i) We derive necessary and sufficient conditions for a distribution shift to reveal a model's reliance on spurious features; when these conditions hold, "accuracy on the line" disappears. (ii) We audit leading OOD datasets and find that most still display accuracy on the line, suggesting they are misspecified for evaluating robustness to spurious correlations. (iii) We catalog the few well-specified datasets and summarize generalizable design principles, such as identifying datasets of natural interventions (e.g., a pandemic), to guide future well-specified benchmarks.
URL: https://openreview.net/forum?id=fNywRyqPQo
---
Title: BiDoRA: Bi-level Optimization-Based Weight-Decomposed Low-Rank Adaptation
Authors: Peijia Qin, Ruiyi Zhang, Pengtao Xie
Abstract: Parameter-efficient fine-tuning (PEFT) is a flexible and efficient method for adapting large language models (LLMs) to downstream tasks. Among these methods, weight-decomposed low-rank adaptation (DoRA) is a promising approach that decomposes weight matrices into magnitude and direction components to mimic full fine-tuning (FT) better. However, DoRA's simultaneous optimization of these components makes it over-expressive, increases the risk of overfitting, and creates a coupled updating pattern that limits its learning capacity. To address these issues, we propose Bi-level Optimization-Based Weight-Decomposed Low-Rank Adaptation (BiDoRA), a novel PEFT method based on a bi-level optimization framework. BiDoRA fundamentally differs from DoRA by optimizing the magnitude and direction in two separate, asynchronous loops using distinct training and validation data splits. This decoupled optimization process effectively mitigates overfitting and allows for more flexible updates that align even more closely with FT. For instance, weight decomposition analysis shows BiDoRA achieves a magnitude-direction update correlation of $-8.042$, significantly closer to the FT ideal compared to $-1.784$ for DoRA. Evaluation of BiDoRA on diverse tasks spanning natural language understanding, generation, token classification, and extremely small biomedical datasets reveals that it consistently outperforms DoRA and a wide range of leading PEFT methods. This improvement is statistically significant, as demonstrated on the GLUE benchmark where BiDoRA surpasses DoRA with a p-value of $2.4\times10^{-4}$ in terms of the Wilcoxon signed-rank test. The code for BiDoRA is available at https://github.com/t2ance/BiDoRA.
URL: https://openreview.net/forum?id=v2xCm3VYl4
---
Title: Foldable SuperNets: Scalable Merging of Transformers with Different Initializations and Tasks
Authors: Edan kinderman, Itay Hubara, Haggai Maron, Daniel Soudry
Abstract: Recent methods aim to merge neural networks (NNs) with identical architectures trained on different tasks into a single multi-task model. While most works focus on the simpler setup of merging NNs initialized from a common pre-trained network, we target the harder problem of merging large transformers trained on different tasks from distinct initializations. We show that traditional merging methods fail catastrophically in this setup, while Knowledge Distillation (KD) achieves much better results, though at a higher cost. However, KD is data-inefficient, as it does not exploit the original models' weights. To solve this, we introduce "Foldable SuperNet Merge" (FS-Merge), which trains a SuperNet containing the original models (with frozen weights) using a feature reconstruction objective. After training, the SuperNet is folded back to the size of a single original model. FS-Merge is simple, data-efficient, has a computational cost comparable to KD, and is proven to have superior expressiveness over traditional merging methods. It achieves SOTA results when tested on MLPs and transformers across various sizes, tasks, modalities, and distribution shifts, especially in low-data scenarios.
URL: https://openreview.net/forum?id=6FqwLestHv
---
Title: Random Erasing vs. Model Inversion: A Promising Defense or a False Hope?
Authors: Viet-Hung Tran, Ngoc-Bao Nguyen, Son T. Mai, Hans Vandierendonck, Ira Assent, Alex Kot, Ngai-Man Cheung
Abstract: Model Inversion (MI) attacks pose a significant privacy threat by reconstructing private
training data from machine learning models. While existing defenses primarily concentrate on model-centric approaches, the impact of data on MI robustness remains largely unexplored. In this work, we explore Random Erasing (RE)—a technique traditionally used for improving model generalization under occlusion—and uncover its surprising effectiveness as a defense against MI attacks.
Specifically, our novel feature space analysis shows that model trained with RE-images introduces a significant discrepancy between the features of MI-reconstructed images and those of the private data. At the same time, features of private images remain distinct from other
classes and well-separated from different classification regions. These effects collectively de grade MI reconstruction quality and attack accuracy while maintaining reasonable natural accuracy. Furthermore, we explore two critical properties of RE including Partial Erasure and Random Location. First, Partial Erasure prevents the model from observing entire objects during training, and we find that this has significant impact on MI, which aims to reconstruct the entire objects. Second, the Random Location of erasure plays a crucial role in achieving a strong privacy-utility trade-off. Our findings highlight RE as a simple yet effective defense mechanism that can be easily integrated with existing privacy-preserving techniques. Extensive experiments of 37 setups demonstrate that our method achieves SOTA performance in privacy-utility tradeoff. The results consistently demonstrate the superiority of our defense over existing defenses across different MI attacks, network architectures, and attack configurations. For the first time, we achieve significant degrade in attack accuracy without decrease in utility for some configurations. Our code and additional results are available at: https://ngoc-nguyen-0.github.io/MIDRE/
URL: https://openreview.net/forum?id=S9CwKnPHaO
---
Title: Classifier-Free Guidance is a Predictor-Corrector
Authors: Arwen Bradley, Preetum Nakkiran
Abstract: We investigate the theoretical foundations of classifier-free guidance (CFG). CFG is the dominant method of conditional sampling for text-to-image diffusion models, yet unlike other aspects of diffusion, it remains on shaky theoretical footing. In this paper, we first show that CFG interacts differently with DDPM (Ho et al., 2020) and DDIM (Song et al., 2021), and neither sampler with CFG generates the gamma-powered distribution $p(x|c)^\gamma p(x)^{1-\gamma}$. Then, we clarify the behavior of CFG by showing that it is a kind of predictor-corrector method (Song et al., 2020) that alternates between denoising and sharpening, which we call predictor-corrector guidance (PCG). We prove that in the SDE limit, CFG is actually equivalent to combining a DDIM predictor for the conditional distribution together with a Langevin dynamics corrector for a gamma-powered distribution (with a carefully chosen gamma). Our work thus provides a lens to theoretically understand CFG by embedding it in a broader design space of principled sampling methods.
URL: https://openreview.net/forum?id=zrWNtzSZsf
---
Title: Semantic Mapping in Indoor Embodied AI - A Survey on Advances, Challenges, and Future Directions
Authors: Sonia Raychaudhuri, Angel X Chang
Abstract: Intelligent embodied agents (e.g. robots) need to perform complex semantic tasks in unfamiliar environments. Among many skills that the agents need to possess, building and maintaining a semantic map of the environment is most crucial in long-horizon tasks. A semantic map captures information about the environment in a structured way, allowing the agent to reference it for advanced reasoning throughout the task. While existing surveys in embodied AI focus on general advancements or specific tasks like navigation and manipulation, this paper provides a comprehensive review of semantic map-building approaches in embodied AI, specifically for indoor navigation. We categorize these approaches based on their structural representation (spatial grids, topological graphs, dense point-clouds or hybrid maps) and the type of information they encode (implicit features or explicit environmental data). We also explore the strengths and limitations of the map building techniques, highlight current challenges, and propose future research directions. We identify that the field is moving towards developing open-vocabulary, queryable, task-agnostic map representations, while high memory demands and computational inefficiency still remaining to be open challenges. This survey aims to guide current and future researchers in advancing semantic mapping techniques for embodied AI systems.
URL: https://openreview.net/forum?id=USgQ38RG6G
---
Title: Joint Generative Modeling of Grounded Scene Graphs and Images via Diffusion Models
Authors: Bicheng Xu, Qi Yan, Renjie Liao, Lele Wang, Leonid Sigal
Abstract: A grounded scene graph represents a visual scene as a graph, where nodes denote objects (including labels and spatial locations) and directed edges encode relations among them. In this paper, we introduce a novel framework for joint grounded scene graph - image generation, a challenging task involving high-dimensional, multi-modal structured data. To effectively model this complex joint distribution, we adopt a factorized approach: first generating a grounded scene graph, followed by image generation conditioned on the generated grounded scene graph. While conditional image generation has been widely explored in the literature, our primary focus is on the generation of grounded scene graphs from noise, which provides efficient and interpretable control over the image generation process. This task requires generating plausible grounded scene graphs with heterogeneous attributes for both nodes (objects) and edges (relations among objects), encompassing continuous attributes (e.g., object bounding boxes) and discrete attributes (e.g., object and relation categories). To address this challenge, we introduce DiffuseSG, a novel diffusion model that jointly models the heterogeneous node and edge attributes. We explore different encoding strategies to effectively handle the categorical data. Leveraging a graph transformer as the denoiser, DiffuseSG progressively refines grounded scene graph representations in a continuous space before discretizing them to generate structured outputs. Additionally, we introduce an IoU-based regularization term to enhance empirical performance. Our model outperforms existing methods in grounded scene graph generation on the Visual Genome and COCO-Stuff datasets, excelling in both standard and newly introduced metrics that more accurately capture the task’s complexity. Furthermore, we demonstrate the broader applicability of DiffuseSG in two important downstream tasks: (1) achieving superior results in a range of grounded scene graph completion tasks, and (2) enhancing grounded scene graph detection models by leveraging additional training samples generated by DiffuseSG. Code is available at https://github.com/ubc-vision/DiffuseSG.
URL: https://openreview.net/forum?id=2cxxZI2LOL
---
New submissions
===============
Title: LLM4FL: Multi-Agent Repository-Level Software Fault Localization via Graph-Based Retrieval and Iterative Refinement
Abstract: Locating and fixing software faults is a time-consuming and resource-intensive task in software development. Traditional fault localization methods, such as Spectrum-Based Fault Localization (SBFL), rely on statistical analysis of test coverage data but often lack accuracy. While more effective, learning-based techniques require large training datasets and can be computationally intensive. Recent advancements in Large Language Models (LLMs) have shown potential for improving fault localization by enhancing code comprehension and reasoning. LLMs are typically pretrained and can be leveraged for fault localization without additional training. However, these LLM-based techniques face challenges, including token limitations, performance degradation with long inputs, and difficulties managing large-scale projects with complex, interacting components. We introduce LLM4FL, a multi-LLM-agent-based fault localization approach to address these challenges. LLM4FL utilizes three agents. First, the Context Extraction Agent uses an order-aware division strategy to divide and analyze extensive coverage data into small groups within the LLM's token limit, identify the failure reason, and prioritize failure-related methods. The prioritized methods are sent to the Debugger Agent, which uses graph-based retrieval to identify failure reasons and rank suspicious methods in the codebase. Then the Reviewer Agent re-evaluates and re-ranks buggy methods using verbal reinforcement learning and self-criticism. Evaluated on the Defects4J (V2.0.0) benchmark of 675 faults from 14 Java projects, LLM4FL outperforms AutoFL by 18.55% in Top-1 accuracy and surpasses supervised methods like DeepFL and Grace, all without task-specific training. Coverage splitting and prompt chaining further improve performance, boosting Top-1 accuracy by up to 22%.
URL: https://openreview.net/forum?id=z91EvZbSI1
---
Title: Mitigating Steady-State Bias in Off-Policy TD Learning via Distributional Correction
Abstract: We explore the off-policy value prediction problem in the reinforcement learning setting, where one estimates the value function of the target policy using the sample trajectories obtained from a behaviour policy. Applying importance sampling based methods are typically a go-to approach for getting such estimates but tend to suffer high error in long-horizon problems since it can only correct single-step discrepancies and fails to address steady-state bias - skewed state visitation under the behavior policy. In this paper,
we present an algorithm for alleviating this bias in the off-policy value prediction using linear function approximation by correcting the state visitation distribution discrepancies. We establish rigorous theoretical guarantees, proving asymptotic convergence under Markov noise with ergodicity and demonstrating that the spectral properties of the corrected update matrix ensure stability. Most significantly, we derive an error decomposition showing that the total estimation error is bounded by a constant multiple of the best achievable approximation within the function class, where this constant transparently depends on distribution estimation quality and feature design. Empirical evaluation across multiple benchmark domains demonstrates that our method effectively mitigates steady-state bias and can be a viable alternative to existing methods in scenarios where distributional shift is critical.
URL: https://openreview.net/forum?id=QLZAHgiowr
---
Title: Measuring Superposition with Sparse Autoencoders — Does Superposition Cause Adversarial Vulnerability?
Abstract: Neural networks achieve remarkable performance through \textit{superposition}—encoding multiple features as overlapping directions in activation space rather than dedicating individual neurons to each feature. This phenomenon fundamentally challenges interpretability: when neurons respond to multiple unrelated concepts, understanding network behavior becomes intractable. Yet despite its central importance, we lack principled methods to measure superposition. We present an information-theoretic framework that measures the effective number of features through the exponential of Shannon entropy applied to sparse autoencoder activations. This threshold-free metric, grounded in rate-distortion theory and analogy to quantum entanglement, provides the first universal measure of superposition applicable to any neural network.
Our approach demonstrates strong empirical validation: correlation with ground truth exceeds 0.94 in toy models, accurately detects minimal superposition in algorithmic tasks (feature count approximately equals neuron count), and reveals systematic feature reduction under capacity constraints (up to 50\% reduction with dropout). Layer-wise analysis of Pythia-70M reveals feature counts peak in early-middle layers at 20 times the number of neurons before declining—mirroring patterns observed in intrinsic dimensionality studies. The metric also captures developmental dynamics, detecting sharp reorganization during grokking phase transitions where models shift from superposed memorization to compact algorithmic solutions.
Surprisingly, adversarial training can increase feature counts by up to 4× while improving robustness, contradicting the hypothesis that superposition causes vulnerability. The effect depends on task complexity and network capacity: simple tasks and ample capacity enable feature expansion, while complex tasks or limited capacity force feature reduction.
By providing a principled, threshold-free measure of superposition, this work enables quantitative study of neural information organization.
URL: https://openreview.net/forum?id=qaNP6o5qvJ
---
Title: Beyond Naïve Prompting: Strategies for Improved Zero-shot Context-aided Forecasting with LLMs
Abstract: Forecasting in real-world settings requires models to integrate not only historical data but also relevant contextual information, often available in textual form. While recent work has shown that large language models (LLMs) can be effective context-aided forecasters via naïve direct prompting, their full potential remains underexplored. We address this gap with 4 strategies, providing new insights into the zero-shot capabilities of LLMs in this setting. ReDP improves interpretability by eliciting explicit reasoning traces, allowing us to assess the model’s reasoning over the context independently from its forecast accuracy. CorDP leverages LLMs solely to refine existing forecasts with context, enhancing their applicability in real-world forecasting pipelines. IC-DP proposes embedding historical examples of context-aided forecasting tasks in the prompt, substantially improving accuracy even for the largest models. Finally, RouteDP optimizes resource efficiency by using LLMs to estimate task difficulty, and routing the most challenging tasks to larger models. Evaluated on different kinds of context-aided forecasting tasks from the CiK benchmark, our strategies demonstrate distinct benefits over naïve prompting across LLMs of different sizes and families. These results open the door to further simple yet effective improvements in LLM-based context-aided forecasting.
URL: https://openreview.net/forum?id=CLuE0GL5xE
---
Title: Amortized Bayesian Workflow
Abstract: Bayesian inference often faces a trade-off between computational speed and sampling accuracy. We propose an adaptive workflow that integrates rapid amortized inference with gold-standard MCMC techniques to achieve a favorable combination of both speed and accuracy when performing inference on many observed datasets. Our approach uses principled diagnostics to guide the choice of inference method for each dataset, moving along the Pareto front from fast amortized sampling via generative neural networks to slower but guaranteed-accurate MCMC when needed. By reusing computations across steps, our workflow synergizes amortized and MCMC-based inference. We demonstrate the effectiveness of this integrated approach on several synthetic and real-world problems with tens of thousands of datasets, showing efficiency gains while maintaining high posterior quality.
URL: https://openreview.net/forum?id=osV7adJlKD
---
Title: Eigenvector Phase Transitions under Anisotropic Noise
Abstract: Identifying latent structures in environmental data—such as habitat clusters or pollution sources—is a fundamental challenge in ecological and climate science. Spectral methods, which analyse the principal eigenvectors of affinity matrices, are powerful tools for this task. However, environmental systems are rarely isotropic; physical processes like river flows or prevailing winds create strong directional gradients, resulting in anisotropic noise. The effect of such anisotropy on the reliability of spectral methods is not yet well understood in the literature. In this work, we develop a rigorous theory for this scenario by analysing a spiked random matrix model subjected to anisotropic noise. We derive an exact, analytical expression for the critical signal-to-noise ratio required for signal detection, establishing a sharp phase transition. Our central result proves that this threshold depends critically on the geometric alignment between the signal and the dominant environmental gradient, formalising a ''camouflage effect''. We also uncover a critical failure mode where this environmental gradient can itself create a ''phantom'' structure that spectral methods can easily detect, posing a significant risk of misinterpretation for scientists. Furthermore, we show that in the detectable phase, the second eigenvector aligns with the primary noise direction, revealing a deeper reorganisation of the system's structure. We complete our analysis with a Central Limit Theorem for the alignment fluctuations. We validate our theoretical predictions with simulations of ecological systems, offering a fundamental understanding of when spectral methods succeed or fail in realistic environments. Code to reproduce all results in the paper is anonymously released at https://anonymous.4open.science/r/tmlr_ept
URL: https://openreview.net/forum?id=X1EqLH399m
---
Title: Decentralized Policy Gradients for Optimizing Generalizable Policies in Multi-Agent Reinforcement Learning
Abstract: Parameter Sharing (PS) is a widely used practice in Multi-Agent Reinforcement Learning (MARL), where a single neural network is shared among all agents. Despite its efficiency and effectiveness, PS can occasionally result in suboptimal performance. While prior research has primarily addressed this issue from the perspective of update conflicts among different agents, we investigate it from an optimization standpoint. Specifically, we point out the analogy between PS in MARL and Centralized SGD (CSGD) in distributed learning and hypothesize that PS may inherit similar convergence and generalization issues as CSGD, such as lower convergence levels of key metrics and larger generalization gaps. To address these issues, we propose Decentralized Policy Gradients (DecPG), which leverages the principles of Decentralized SGD. We use an environment with additional noise injected into the observation and action spaces to evaluate the generalization of DecPG. Empirical results show that DecPG outperforms its centralized counterpart, PS, across various aspects---achieving higher rewards, smaller generalization gaps, and flatter reward landscapes. The results confirm that PS suffers from convergence and generalization issues similar to those of CSGD, and show that our DSGD-based method, DecPG, effectively mitigates these problems---offering a new optimization perspective on MARL algorithm performance.
URL: https://openreview.net/forum?id=utpzisYFqd
---
Title: Concept Siever : Towards Controllable Erasure of Concepts from Diffusion Models without Side-effect
Abstract: Diffusion models' unprecedented success with image generation can largely be attributed to their large-scale pretraining on massive datasets. Yet, the necessity of forgetting specific concepts for regulatory or copyright compliance poses a critical challenge. Existing approaches in concept forgetting, although reasonably successful in forgetting a given concept, frequently fail to preserve generation quality or demand extensive domain expertise for preservation. To alleviate such issues, we introduce Concept Siever, an end-to-end framework for targeted concept removal within pre-trained text-to-image diffusion models. The foundation of Concept Siever rests on \textit{two key innovations}: First, an automatic technique to create paired dataset of target concept and its negations by utilizing the diffusion model’s latent space. A key property of these pairs is that they differ only in the target concept, enabling forgetting with \textit{minimal side effects} and \textit{without requiring domain expertise}. Second, we present Concept Sieve, a localization method for identifying and isolating the model components most responsible to the target concept. By retraining only these localized components on our paired dataset for a target concept, Concept Siever accurately removes the concept with \textit{negligible side-effects, preserving neighboring and unrelated concepts}. Moreover, given the subjective nature of forgetting a concept like nudity, we propose Concept Sieve which provides a \texit{fine-grained control over the forgetting strength at inference time}, catering to diverse deployment needs without any need of finetuning. We report state-of-the-art performance on the I2P benchmark, surpassing previous domain-agnostic methods by over $33\%$ while showing superior structure preservation. We validate our results through extensive quantitative and qualitative evaluation along with a user study.
URL: https://openreview.net/forum?id=O7zTvlSBZ9
---
Title: Tex4D: Zero-shot 4D Character Texturing with Video Diffusion Models
Abstract: 3D meshes are widely used in movies, games, AR, and VR for their efficiency in animation and minimal memory footprint, leading to the creation of a large number of mesh sequences. However, creating dynamic textures for these mesh sequences to model the appearance transformations remains labor-intensive for professional artists. In this work, we present Tex4D, a zero-shot approach that creates multi-view and temporally consistent dynamic mesh textures by integrating the inherent 3D geometry knowledge with the expressiveness of video diffusion models. Given an untextured mesh sequence and a text prompt as inputs, our method enhances multi-view consistency by synchronizing the diffusion process across different views through latent aggregation in the UV space. To ensure temporal consistency, such as lighting changes, wrinkles, and appearance transformations, we leverage prior knowledge from a conditional video generation model for texture synthesis. Using the video diffusion model and the UV texture aggregation in a straightforward way leads to blurred results. We analyze the underlying causes and propose a simple yet effective modification to the DDIM sampling process to address this issue. Additionally, we introduce a reference latent texture to strengthen the correlation between frames during the denoising process. To the best of our knowledge, Tex4D is the first method specifically designed for 4D character texturing. Extensive experiments demonstrate its superiority in producing multi-view and multi-frame consistent dynamic textures for mesh sequences.
URL: https://openreview.net/forum?id=y1EBrNkzDa
---
Title: Robust Clustering using Gaussian Mixtures in the Presence of Cellwise Outliers
Abstract: In this paper we propose a novel algorithm for robust estimation of Gaussian Mixture Model (GMM) parameters and clustering that explicitly accounts for cell outliers. To achieve this, the proposed algorithm minimizes a penalized negative log-likelihood function where the penalty term is derived via the false discovery rate principle. The penalized negative log-likelihood function is cyclically minimized over outlier positions and the GMM parameters. Furthermore, the minimization over the GMM parameters is done using the majorization minimization framework: specifically we minimize a tight upper bound on the negative log-likelihood function which decouples into simpler optimization subproblems that can be solved efficiently.
We present several numerical simulation studies comprising experiments aimed at evaluating the performance of the proposed method on synthetic as well as real world data and at systematically comparing it with state-of-the-art robust techniques in different scenarios. The simulation studies demonstrate that our approach effectively addresses the challenges inherent in parameter estimation of GMM and clustering in contaminated data environments.
URL: https://openreview.net/forum?id=oVHPEgjdWk
---
Title: Adversarial Fine-tuning of Compressed Neural Networks for Joint Improvement of Robustness and Efficiency
Abstract: As deep learning (DL) models are increasingly being integrated into our everyday lives, ensuring their safety by making them robust against adversarial attacks has become increasingly critical. DL models have been found to be susceptible to adversarial attacks by introducing small, targeted perturbations to disrupt the input data. Adversarial training has been presented as a mitigation strategy that can result in more robust models. This adversarial robustness comes with additional computational costs required to design adversarial attacks during training. The two objectives -- adversarial robustness and computational efficiency -- then appear to be in conflict with each other. In this work, we explore the effects of neural network compression on adversarial robustness. We specifically explore the effects of fine-tuning on compressed models, and present the trade-off between standard fine-tuning and adversarial fine-tuning. Our results show that {\em adversarial fine-tuning} of compressed models can yield large improvements to their robustness performance. We present experiments on several benchmark datasets showing that adversarial fine-tuning of compressed models can achieve robustness performance comparable to adversarially trained models, while also improving computational efficiency.
URL: https://openreview.net/forum?id=a1EIdh3RSD
---
Title: Improving Adversarial Training for Two-player Competitive Games via Episodic Reward Engineering
Abstract: Training adversarial agents to attack neural network policies has proven to be both effective and practical. However, we observe that existing methods can be further enhanced by distinguishing between states leading to win or lose and encouraging the policy training by reward engineering to prioritize winning states. In this paper, we introduce a novel adversarial training method with reward engineering for two-player competitive games. Our method extracts the historical evaluations for states from historical experiences with an episodic memory, and then incorporating these evaluations into the rewards with our proposed reward revision method to improve the adversarial policy optimization. We evaluate our approach using two-player competitive games in MuJoCo simulation environments, demonstrating that our method establishes the most promising attack performance and defense difficulty against the victims among the existing adversarial policy training techniques.
URL: https://openreview.net/forum?id=z4XtJWJC9K
---
Title: Architecture-Aware Generalization Bounds for Temporal Networks: Theory and Fair Comparison Methodology
Abstract: Deep temporal architectures such as Temporal Convolutional Networks (TCNs) achieve strong predictive performance on sequential data, yet theoretical understanding of their generalization remains limited. We address this gap by providing both the first non-vacuous, architecture-aware generalization bounds for deep temporal models and a principled evaluation methodology.
For exponentially $\beta$-mixing sequences, we derive bounds scaling as
$
\mathcal{O}\!\Bigl(R\,\sqrt{\tfrac{D\,p\,n\,\log N}{N}}\Bigr),
$
where $D$ is network depth, $p$ kernel size, $n$ input dimension, and $R$ weight norm. Our delayed-feedback blocking mechanism transforms dependent samples into effectively independent ones while discarding only $O(1/\log N)$ of the data, yielding $\sqrt{D}$ scaling instead of exponential-implying that doubling depth requires approximately quadrupling the training data.
We also introduce a fair-comparison methodology that fixes the effective sample size to isolate the effect of temporal structure from information content. Under $N_{\text{eff}}=2{,}000$, strongly dependent sequences ($\rho=0.8$) exhibit $\approx76\%$ smaller generalization gaps than weakly dependent ones ($\rho=0.2$), challenging the intuition that dependence is purely detrimental. Yet convergence rates diverge from theory: weak dependencies follow $N_{\text{eff}}^{-1.21}$ scaling and strong dependencies follow $N_{\text{eff}}^{-0.89}$, both steeper than the predicted $N^{-0.5}$. These findings reveal that temporal dependence can enhance learning under fixed information budgets, while highlighting gaps between theory and practice that motivate future research.
URL: https://openreview.net/forum?id=7GZ0TcV691
---
Title: Multimodal Cultural Safety: Evaluation Framework and Alignment Strategies
Abstract: Large vision-language models (LVLMs) are increasingly deployed in globally distributed applications, such as tourism assistants, yet their ability to produce culturally appropriate responses remains underexplored. Existing multimodal safety benchmarks primarily focus on physical safety and overlook violations rooted in cultural norms, which can result in symbolic harm. For example, suggesting clocks as gifts for a baby’s birthday in China may invoke associations with death, leading to user discomfort and undermining trust. To address this gap, we introduce CROSS, a benchmark designed to assess the cultural safety reasoning capabilities of LVLMs. CROSS includes 1,284 multilingual visually grounded queries from 16 countries, three everyday domains (i.e., shopping, meal planning, and outdoor activities), and 14 languages, where cultural norm violations emerge only when images are interpreted in context. We propose CROSS-Eval, an intercultural theory-based framework that measures four key dimensions: cultural awareness, norm education, compliance, and helpfulness. Using this framework, we evaluate 21 leading LVLMs, including mixture-of-experts models (e.g., Llama-4-Maverick) and reasoning models (e.g., o1 and Gemini-2.5-Pro). Results reveal significant cultural safety gaps: the best-performing model achieves only 61.79% in awareness and 37.73% in compliance. While some open-source models achieve performance better or comparable to GPT-4o, they still fall notably short of proprietary models. Our results further show that increasing reasoning capacity improves cultural alignment but does not fully resolve the issue. To improve model performance, we develop two enhancement strategies: supervised fine-tuning with culturally grounded, open-ended data and preference tuning with contrastive response pairs that highlight safe versus unsafe behaviors. These methods substantially improve GPT-4o’s cultural awareness (+60.14%) and compliance (+55.2%), while preserving general multimodal capabilities with minimal performance reduction on general multimodal understanding benchmarks. This work establishes a framework for evaluating and improving cultural safety in vision-language systems across diverse global contexts.
URL: https://openreview.net/forum?id=mkFBmxgnRh
---
Title: CLIP Meets Diffusion: A Synergistic Approach to Anomaly Detection
Abstract: Anomaly detection is a complex problem due to the ambiguity in defining anomalies, the diversity of anomaly types (e.g., local and global defect), and the scarcity of training data. As such, it necessitates a comprehensive model capable of capturing both low-level and high-level features, even with limited data. To address this, we propose CLIPFUSION, a method that leverages both discriminative and generative foundation models. Given the CLIP-based discriminative model's limited capacity to capture fine-grained local details, we incorporate a diffusion-based generative model to complement its features. This integration yields a synergistic solution for anomaly detection. To this end, we propose using diffusion models as feature extractors for anomaly detection, and introduce carefully designed strategies to extract informative cross-attention and feature maps. Experimental results on benchmark datasets (MVTec-AD, VisA) demonstrate that CLIPFUSION consistently outperforms baseline methods in both anomaly segmentation and classification under both zero-shot and few-shot settings. We believe that our method underscores the effectiveness of multi-modal and multi-model fusion in tackling the multifaceted challenges of anomaly detection, providing a scalable solution for real-world applications.
URL: https://openreview.net/forum?id=WpFzZNuQmg
---
Title: Oscillations Make Neural Networks Robust to Quantization
Abstract: We challenge the prevailing view that weight oscillations observed during Quantization Aware Training (QAT) are merely undesirable side-effects and argue instead that they are an essential part of QAT. We show in a linear model with a single weight that the straight-through estimator (STE) results in an additional loss term that causes oscillations by pushing weights away from their nearest quantization level. Based on the mechanism from the analysis, we then derive a regularizer that induces oscillations in the weights of neural networks during training. Our empirical results on ResNet-18 and Tiny ViT on CIFAR-10 and Tiny-ImageNet datasets demonstrate across a range of quantization levels that training with oscillations followed by post-training quantization (PTQ) is sufficient to recover the performance of QAT in most cases. With this work we shed further light on the dynamics of QAT and contribute a novel insight into explaining the role of oscillations in QAT which until now have been considered to have a primarily negative effect on quantization.
URL: https://openreview.net/forum?id=bPwcJ0nkDC
---
Title: Conditional Kernel Quantile Embeddings: A Nonparametric Framework for Conditional Two-Sample Testing
Abstract: Comparing conditional probability distributions, P(Y∣X) and Q(Y∣X), is a fundamental problem in machine learning, crucial for tasks like causal inference, detecting dataset shift, and model validation. The predominant approach, based on Conditional Kernel Mean Embeddings (KCMEs), suffers from significant drawbacks: it relies on strong and often unverifiable assumptions on the kernel to be a metric, incurs high computational costs, and may exhibit reduced sensitivity to higher-order distributional differences. We introduce Conditional Kernel Quantile Embeddings (CKQEs), a novel and robust framework for representing conditional distributions in a Reproducing Kernel Hilbert Space (RKHS). Throughout, we assume P_X = Q_X for conditional comparisons, and we require only that the output-space kernel be quantile-characteristic. From CKQEs, we construct the Conditional Kernel Quantile Discrepancy (CKQD), a new family of probability metrics. We prove that CKQD: (1) is a metric under substantially weaker and more practical kernel conditions than KCME-based distances, namely requiring only a quantile-characteristic kernel; (2) possesses a clear geometric interpretation, recovering a conditional version of the Sliced Wasserstein distance in a special case; and (3) admits a computationally efficient, statistically consistent non-parametric estimator with proven finite-sample convergence rates. By addressing the core weaknesses of the KCME framework, CKQE provides a more versatile and theoretically sound foundation for conditional two-sample testing.
URL: https://openreview.net/forum?id=AuqmLks0T7
---
Title: Exploring exploration with foundation agents in interactive environments
Abstract: While foundation models have recently shown exemplary progress solving difficult single-turn math and reasoning problems, many human endeavors---from conducting scientific research to developing new technologies---require multi-turn exploration in dynamic interactive environments. Crucial components of learning from experience in these settings, such as efficiently gathering information to test hypotheses, meta-learning a model of the world's dynamics, and adapting to unexpected changes, remain largely unexplored for these models. We first evaluate foundation models in Feature World, a setting that primarily tests information gathering about a static hidden reward function. In this initial setting, we show that state-of-the-art foundation models come close to optimal efficiency in selecting maximally informative actions in tasks with simple reward functions, with more recent and thinking models performing especially well. As a proof of concept, we also show a model can gather information efficiently in a 3D embodied version of this task, though errors in vision limit some aspects of performance. In order to test exploration across multiple dependent turns and trials, we implement a custom, text-based version of the Alchemy environment, a benchmark designed for meta-learning. Here, agents must deduce a latent causal structure governing object interactions by integrating information gathered over a sequence of trials where actions modify the state relevant to future outcomes. In this more complex setting, we find that recent foundation models struggle to meta-learn strategies that enable improved performance over time. However, prompting the models to summarize their observations at regular intervals enables an emergent meta-learning process, allowing them to improve across trials. Notably, in some models, summarization also enabled adaptive re-learning of this information when the environment's rules change unexpectedly. While most models performed reasonably well on simple Feature World tasks, evaluations in Alchemy reveal stark differences in robustness among the models, with Gemini 2.5 performing best, followed by Claude 3.7, and ChatGPT-4o and o4-mini struggling the most. These results underscore Alchemy's value as a benchmark for meta-learning and strategy adaptation in foundation models. By moving beyond simple discovery to complex, stateful environments, we demonstrate that the most significant challenge for foundation agents is not selecting informative actions in the moment, but rather seeking and integrating knowledge through adaptive strategies over time. Intriguingly, we find there is likely no intrinsic barrier to future generations of foundation agents more fully mastering these abilities.
URL: https://openreview.net/forum?id=wOrkUTr0W5
---
Title: Melody or Machine: Detecting Synthetic Music with Dual- Stream Contrastive Learning
Abstract: The rapid evolution of end-to-end AI music generation poses an escalating threat to artistic authenticity and copyright, demanding detection methods that can keep pace. While foundational, existing models like SpecTTTra falter when faced with the diverse and rapidly advancing ecosystem of new generators, exhibiting significant performance drops on out-of-distribution (OOD) content. This generalization failure highlights a critical gap: the need for more challenging benchmarks and more robust detection architectures. To address this, we first introduce Melody or Machine (MoM), a new large-scale benchmark of over 130,000 songs (6,665 hours). MoM is the most diverse dataset to date, built with a mix of open and closed-source models and a curated OOD test set designed specifically to foster the development of truly generalizable detectors. Alongside this benchmark, we introduce CLAM, a novel dual-stream detection architecture. We hypothesize that subtle, machine-induced inconsistencies between vocal and instrumental elements, often imperceptible in a mixed signal, offer a powerful tell-tale sign of synthesis. CLAM is designed to test this hypothesis by employing two distinct pre-trained audio encoders (MERT and Wave2Vec2) to create parallel representations of the audio. These representations are fused by a learnable cross-aggregation module that models their inter-dependencies. The model is trained with a dual-loss objective: a standard binary cross-entropy loss for classification, complemented by a contrastive triplet loss which trains the model to distinguish between coherent and artificially mismatched stream pairings, enhancing its sensitivity to synthetic artifacts without presuming a simple feature alignment. CLAM establishes a new state-of-the-art in synthetic music forensics. It achieves an F1 score of 0.925 on our challenging MoM benchmark, significantly outperforming the previous SOTA's 0.869 on the same dataset. This result demonstrates superior generalization to unseen generative models. Furthermore, CLAM scores 0.993 on the popular SONICS benchmark, confirming its effectiveness and setting a new performance standard.
URL: https://openreview.net/forum?id=Ufwes0o2e3
---
Title: Towards Online Multimodal Social Interaction Understanding
Abstract: In this paper, we introduce a new problem, Online-MMSI, where the model must perform multimodal social interaction understanding (MMSI) using only historical information. Given a recorded video and a multi-party dialogue, the AI assistant is required to immediately identify the speaker’s referent, which is critical for real-world human-AI interaction. Without access to future conversational context, both humans and models experience substantial performance degradation when moving from offline to online settings.
\
To tackle the challenges, we propose Online-MMSI-VLM, a novel framework based on multimodal large language models. The core innovations of our approach lie in two components: (1) multi-party conversation forecasting, which predicts upcoming speaker turns and utterances in a coarse-to-fine manner; and (2) socially-aware visual prompting, which highlights salient social cues in each video frame using bounding boxes and body keypoints.
\
Our model achieves state-of-the-art results on three tasks across two datasets, significantly outperforming the baseline and demonstrating the effectiveness of Online-MMSI-VLM.
URL: https://openreview.net/forum?id=5P7yVfUEuD
---
Title: Overcoming Non-stationary Dynamics with Evidential Proximal Policy Optimization
Abstract: Continuous control of non-stationary environments is a major challenge for deep reinforcement learning algorithms. The time-dependency of the state transition dynamics aggravates the notorious stability problems of model-free deep actor-critic architectures. We posit that two properties will play a key role in overcoming non-stationarity in transition dynamics: (i) preserving the plasticity of the critic network and (ii) directed exploration for rapid adaptation to the changing dynamics. We show that performing on-policy reinforcement learning with an evidential critic provides both. The evidential design ensures a fast and sufficiently accurate approximation to the uncertainty around the state-value, which maintains the plasticity of the critic network by detecting the distributional shifts caused by the change in dynamics. The probabilistic critic also makes the actor training objective a random variable, enabling the use of directed exploration approaches as a by-product. We name the resulting algorithm \emph{Evidential Proximal Policy Optimization (EPPO)} due to the integral role of evidential uncertainty quantification in both policy evaluation and policy improvement stages. Through experiments on non-stationary continuous control tasks, where the environment dynamics change at regular intervals, we demonstrate that our algorithm outperforms state-of-the-art on-policy reinforcement learning variants in both task-specific and overall return.
URL: https://openreview.net/forum?id=KTfTwxsVNE
---
Title: Nonlinear reconciliation: Error reduction theorems
Abstract: Forecast reconciliation, an ex-post technique applied to forecasts that must satisfy constraints, has been a prominent topic in the forecasting literature over the past two decades. Recently, several efforts have sought to extend reconciliation methods to the probabilistic settings. Nevertheless, formal theorems demonstrating error reduction in nonlinear contexts, analogous to those presented in Panagiotelis et al., (2021), are still lacking. This paper addresses that gap by establishing such theorems for various classes of nonlinear hypersurfaces and vector-valued functions. Specifically, we derive an exact analog of Theorem 3.1 from Panagiotelis et al., (2021) for hypersurfaces with constant-sign curvature. Additionally, we provide an error reduction theorem for the broader case of hypersurfaces with non-constant-sign curvature and for general manifolds with codimension > 1. To support reproducibility and practical adoption, we release a JAX-based Python package, \emph{to be released upon publication}, implementing the presented theorems and reconciliation procedures.
URL: https://openreview.net/forum?id=dXRWuogm3J
---
Title: TimeAutoDiff: A Unified Framework for Generation, Imputation, Forecasting, and Time-Varying Metadata Conditioning of Heterogeneous Time Series Tabular Data
Abstract: We present \texttt{TimeAutoDiff}, a unified latent–diffusion framework that addresses four fundamental time-series tasks—unconditional generation, missing-data imputation, forecasting, and time-varying-metadata conditional generation—within a single model that natively handles heterogeneous features (continuous, binary, and categorical).
We unify these tasks through a simple masked-modeling strategy: a binary mask specifies which time–feature cells are observed and which must be generated.
To make this work on mixed data types, we pair a lightweight variational autoencoder—which maps continuous, categorical, and binary variables into a continuous latent sequence—with a diffusion model that learns dynamics in that latent space, avoiding separate likelihoods for each data type while still capturing temporal and cross-feature structure.
Two design choices give \texttt{TimeAutoDiff} clear speed and scalability advantages.
First, the diffusion process samples a single latent trajectory for the full horizon \(1{:}T\) rather than denoising one timestep at a time; this whole-sequence sampling drastically reduces reverse-diffusion calls and yields an order-of-magnitude throughput gain.
Second, the VAE compresses along the feature axis, so very wide tables are modeled in a lower-dimensional latent space, further reducing computational load.
Across six real-world datasets, \texttt{TimeAutoDiff} matches or surpasses strong baselines in synthetic sequence fidelity (discriminative, temporal-correlation, and predictive metrics) and consistently lowers MAE/MSE for imputation and forecasting tasks.
Time-varying-metadata conditioning unlocks real-world scenario exploration: by editing metadata sequences (e.g., regime labels, environmental or policy indicators), practitioners can generate coherent families of counterfactual trajectories that track intended directional changes, preserve cross-feature dependencies, and remain conditionally calibrated—making ``what-if'' analysis practical.
Ablations attribute performance gains to whole-sequence sampling, latent compression, and mask conditioning, while a distance-to-closest-record audit indicates strong generalization with limited memorization.
Code implementations of \texttt{TimeAutoDiff} are provided in https://anonymous.4open.science/r/TimeAutoDiff-TMLR-7BA8/README.md.
URL: https://openreview.net/forum?id=bkUd1Dg46c
---
Title: Bregman Centroid Guided Cross-Entropy Method
Abstract: The Cross-Entropy Method (CEM) is a widely adopted trajectory optimizer in model-based reinforcement learning (MBRL), but its unimodal sampling strategy often leads to premature convergence in multimodal landscapes. In this work, we propose \textbf{$\mathcal B$regman-$\mathcal C$entroid Guided CEM ($\mathcal{BC}$-EvoCEM)}, a lightweight enhancement to ensemble CEM that leverages \emph{Bregman centroids} for principled information aggregation and diversity control. BC-EvoCEM computes a performance-weighted Bregman centroid across CEM workers and updates the least contributing ones by sampling within a trust region around the centroid. Leveraging the duality between Bregman divergences and exponential family distributions, we show that BC-EvoCEM integrates seamlessly into standard CEM pipelines with negligible overhead. Empirical results on synthetic benchmarks, a cluttered navigation task, full MBRL pipelines, and a real-world quadruped robot demonstrate that BC-EvoCEM enhances both convergence and solution quality, providing a simple yet effective upgrade for CEM.
URL: https://openreview.net/forum?id=7949RzOul6
---
Title: Efficient Few-Shot Continual Learning in Vision-Language Models
Abstract: Vision-language models (VLMs) excel at tasks like visual question answering and image captioning, but their reliance on frozen, pretrained image encoders like CLIP often leads to persistent vision errors that degrade downstream performance. Moreover, real-world deployment demands that VLMs continually adapt to new, scarce data in a few-shot setting without forgetting prior knowledge. To meet these challenges, we introduce LoRSU (Low-Rank Adaptation with Structured Updates), a lightweight and robust technique for few-shot continual learning of VLMs’ image encoders. Our approach leverages theoretical insights to identify and update only the most critical parameters, achieving significant resource efficiency. Specifically, we demonstrate that LoRSU reduces computational overhead by over 25x compared to full VLM updates, without sacrificing performance. In experiments on VQA benchmarks under a few-shot continual learning protocol, LoRSU demonstrates superior scalability, efficiency, and accuracy, offering a practical solution for dynamic, resource-constrained vision-language applications.
URL: https://openreview.net/forum?id=sQ1w92WW0V
---
Title: Algorithmic Recourse in Abnormal Multivariate Time Series
Abstract: Algorithmic recourse provides actionable recommendations to alter unfavorable predictions of machine learning models, enhancing transparency through counterfactual explanations. While significant progress has been made in algorithmic recourse for static data, such as tabular and image data, limited research explores recourse for multivariate time series, particularly for reversing abnormal time series. This paper introduces Recourse in time series Anomaly Detection (RecAD), a framework for addressing anomalies in multivariate time series using backtracking counterfactual reasoning. By modeling the causes of anomalies as external interventions on exogenous variables, RecAD predicts recourse actions to restore normal status as counterfactual explanations, where the recourse function, responsible for generating actions based on observed data, is trained using an end-to-end approach. Experiments on synthetic and real-world datasets demonstrate its effectiveness.
URL: https://openreview.net/forum?id=kzxFc2Suo5
---
Title: Model-Free Learning with Heterogeneous Dynamical Systems: A Federated LQR Approach
Abstract: We study a model-free federated linear quadratic regulator (LQR) problem where M agents with unknown, distinct yet similar dynamics collaboratively learn an optimal policy to minimize an average quadratic cost while keeping their data private. To exploit the similarity of the agents' dynamics, we propose to use federated learning (FL) to allow the agents to periodically communicate with a central server to train policies by leveraging a larger dataset from all the agents. With this setup, we seek to understand the following questions: (i) Is the learned common policy stabilizing for all agents? (ii) How close is the learned common policy to each agent's own optimal policy? (iii) Can each agent learn its own optimal policy faster by leveraging data from all agents? To answer these questions, we propose the federated and model-free algorithm FedLQR. Our analysis overcomes numerous technical challenges, such as heterogeneity in the agents’ dynamics, multiple local updates, and stability concerns. We show that FedLQR produces a common policy that, at each iteration, is stabilizing for all agents. Moreover, we prove that when learning each agent's optimal policy, FedLQR achieves a sample complexity reduction proportional to the number of agents M in a low-heterogeneity regime, compared to the single-agent setting.
URL: https://openreview.net/forum?id=WSRQeCUc3g
---
Title: Uncovering Language Model Processing Strategies with Non-Negative Per-Example Fisher Factorization
Abstract: Understanding the heuristics and algorithms that comprise a model's behavior is important for safe and reliable deployment.
While gradient clustering has been used for this purpose, gradients of a single log probability capture only a slice of the model's behavior, and clustering can only assign a single factor to each behavior.
We introduce NPEFF (Non-Negative Per-Example Fisher Factorization), an interpretability method that overcomes these limitations by decomposing per-example Fisher matrices using a novel decomposition algorithm that learns a set of components represented by learned rank-1 positive semi-definite matrices.
Through a combination of human evaluation and automated analysis, we demonstrate that these NPEFF components correspond to heuristics used by language models on a variety of text processing tasks.
We find that NPEFF excels at decomposing behaviors comprised of multiple factors compared to the baselines of gradient clustering and activation sparse autoencoders.
We also show how NPEFF can be adapted to be more efficient on tasks with few classes.
We further show how to construct parameter perturbations from NPEFF components to selectively disrupt a given component's role in the model's processing.
Along with conducting extensive ablation studies, we include experiments using NPEFF to study in-context learning.
URL: https://openreview.net/forum?id=DUFvZXrQr7
---
Title: When Are Two Scores Better Than One? Investigating Ensembles of Diffusion Models
Abstract: Diffusion models now generate high-quality, diverse samples, with an increasing focus on more powerful models. Although ensembling is a well-known way to improve supervised models, its application to unconditional score-based diffusion models remains largely unexplored. In this work we investigate whether it provides tangible benefits for generative modelling. We find that while ensemble generally improves the score-matching loss and model likelihood, it fails to consistently enhance perceptual quality metrics such as FID. Our study spans across a breadth of aggregation rules using Deep Ensembles, Monte Carlo Dropout, and Random Forests on CIFAR-10, FFHQ, and tabular data. We attempt to explain this discrepancy by investigating possible explanations, such as the link between score estimation and image quality. Finally, we provide theoretical insights into the summing of score models, which shed light not only on ensembling but also on several model composition techniques (e.g. guidance).
URL: https://openreview.net/forum?id=4iRx9b0Csu
---
Title: Training-Conditional Coverage Bounds under Covariate Shift
Abstract: Conformal prediction methodology has recently been extended to the covariate shift setting, where the distribution of covariates differs between training and test data. While existing results ensure that the prediction sets from these methods achieve marginal coverage above a nominal level, their coverage rate conditional on the training dataset—referred to as training-conditional coverage—remains unexplored. In this paper, we address this gap by deriving upper bounds on the tail of the training-conditional coverage distribution, offering probably approximately correct (PAC) guarantees for these methods. Our results characterize the reliability of the prediction sets in terms of the severity of distributional changes and the size of the training dataset.
URL: https://openreview.net/forum?id=F6hHT3qWxT
---
Title: Interpretability of Language Models for Learning Hierarchical Structures
Abstract: Transformer-based language models are effective but complex, and understanding their inner workings is a significant challenge. Previous research has primarily explored how these models handle simple tasks like name copying or selection; we extend this by investigating how they process complex, recursive language structures defined by context-free grammars (CFGs). We introduce a family of synthetic CFGs that produce hierarchical rules, capable of generating lengthy, locally ambiguous sequences that require dynamic programming to parse. Despite this complexity, we show that generative models like GPT can learn these CFG languages and generate valid completions. Analyzing the model's internals, we find that its hidden states linearly encode parse tree structure (via our new probing technique), and attention patterns statistically align with the information flow of dynamic programming-style parsing algorithms. These provide a controlled interpretability setting for understanding how transformers may represent and compute over hierarchical syntax.
URL: https://openreview.net/forum?id=mPQKyzkA1K
---
Title: LZ Penalty: An information-theoretic repetition penalty for autoregressive language models.
Abstract: We introduce the Lempel-Ziv (LZ) penalty, a penalty specialized for reducing degenerate repetitions in autoregressive language models without loss of capability. The penalty is based on the codelengths in the LZ77 universal lossless compression algorithm. Through the lens of the prediction-compression duality, decoding with the LZ penalty has the interpretation of sampling from the residual distribution after removing the information that is highly compressible. We demonstrate the LZ penalty enables open-source reasoning models to operate with greedy decoding without loss of capability and without instances of degenerate repetition. Both the industry-standard frequency penalty and repetition penalty are ineffective, incurring degenerate repetition rates of up to 4% or more.
URL: https://openreview.net/forum?id=vNzPB4YCHj
---
Title: MaPPO: Maximum a Posteriori Preference Optimization with Prior Knowledge
Abstract: As the era of large language models (LLMs) on behalf of users unfolds, Preference Optimization (PO) methods have become a central approach to aligning LLMs with human preferences and improving performance. We propose Maximum a Posteriori Preference Optimization (MaPPO), a framework for learning from preferences that explicitly incorporates prior reward knowledge into the optimization objective. While existing methods such as Direct Preference Optimization (DPO) and its variants treat preference learning as a Maximum Likelihood Estimation (MLE) problem, MaPPO extends this paradigm by integrating prior reward estimates into a principled Maximum a Posteriori (MaP) objective. This not only generalizes DPO and its variants, but also enhances alignment by mitigating the oversimplified binary classification of responses. More importantly, MaPPO introduces no additional hyperparameter, and supports preference optimization in both offline and online settings. In addition, MaPPO can be used as a plugin with consistent improvement on DPO variants, including widely used SimPO, IPO, and CPO. Extensive empirical evaluations of different model sizes and model series on three standard benchmarks, including MT-Bench, AlpacaEval 2.0, and Arena-Hard, demonstrate consistent improvements in alignment performance without sacrificing computational efficiency.
URL: https://openreview.net/forum?id=NKoV9wEiox
---
Title: CANDOR: Counterfactual ANnotated DOubly Robust Off-Policy Evaluation
Abstract: When applying contextual bandit algorithms in high-stakes settings (e.g., medical treatment), practitioners rely on off-policy evaluation (OPE) methods that use historical data to evaluate the behavior of novel policies prior to deployment. Unfortunately, OPE techniques are inherently limited by the breadth of the available data, which may not reflect distribution shifts resulting from the application of a new policy. Recent work attempts to address this challenge by leveraging domain experts to increase dataset coverage by annotating counterfactual samples. However, such annotations are not guaranteed to be free of errors, and incorporating imperfect annotations can lead to worse policy value estimates than not using the annotations at all. To make use of imperfect annotations, we propose a family of OPE estimators based on the doubly robust (DR) principle, which combines importance sampling (IS) with a reward model (direct method, DM) for better statistical guarantees. We introduce three opportunities within the DR estimation framework to incorporate counterfactual annotations. Under mild assumptions, we prove that using annotations within just the DM component yields the most desirable results, providing an unbiased estimator even under noisy annotations. We validate our approaches in several settings, including a real-world medical domain, observing that the theoretical advantages of using annotations within just the DM component hold in practice under realistic conditions. By addressing the challenges posed by imperfect annotations, this work broadens the applicability of OPE methods and facilitates safer and more effective deployment of decision-making systems.
URL: https://openreview.net/forum?id=PDRFappNFQ
---
Title: Towards Fair In-Context Learning with Tabular Foundation Models
Abstract: Transformer-based tabular foundation models have recently demonstrated promising in-context learning (ICL) performance on structured data, emerging as competitive alternatives to gradient-boosted trees. However, the fairness implications of this new paradigm remain largely unexplored. We present the first investigation of fairness in tabular ICL, evaluating three recently proposed foundation models—TabPFNv2, TabICL, and TabDPT—on multiple benchmark datasets. To mitigate biases, we explore three pre-processing fairness-enhancing methods: correlation removal (decorrelating input features from the sensitive attribute), group-balanced sample selection (ensuring equal representation of protected groups in context examples), and uncertainty-based sample selection (prioritizing context examples with high sensitive-attribute prediction uncertainty). Our experiments show that the uncertainty-based strategy consistently improves group fairness metrics (e.g., demographic parity, equalized odds, and equal opportunity) with minimal impact on predictive accuracy. We release our code to facilitate reproducibility (https://anonymous.4open.science/r/Fair-TabICL-Anonymized)
URL: https://openreview.net/forum?id=AsBhwD0sqo
---
Title: Rethinking Industrial Anomaly Detection in the Era of Large Vision-Language Models
Abstract: State-of-the-art methods for industrial anomaly detection (IAD) typically rely on a training set of images to define normal conditions, flagging any deviations as anomalies. Obtaining this training set has two main issues - it is time consuming to obtain an extensive labeled set, and the assumption that all patterns outside the training set are truly anomalous is often unrealistic. Many rare patterns not captured in the training set, such as environmental changes, positional changes, or permissible deformation, may not constitute actual industrial defects. In this paper, we reframe the IAD task by using large vision-language models (LVLMs) without fine-tuning on training images. LVLMs can interpret and generalize from a single reference image, and can be more robust to rare but acceptable changes in images. Our experiments on two popular benchmarks, MvTec-AD and VisA, show that LVLMs with just one image and a textual description is competitive with state-of-the-art models, and offer a more robust and generalizable solution even with variations in testing images. We also identify a key limitation: LVLM performance degrades when detecting small anomalies. Despite this, our findings highlight the potential of LVLMs as a flexible and scalable foundation for industrial anomaly detection, opening new directions for future research.
URL: https://openreview.net/forum?id=EyhzBqeRNb
---
Title: Trading-off Statistical and Computational Efficiency via $W$-step MDPs: A Policy Gradient Approach
Abstract: In reinforcement learning, algorithm performance is typically evaluated along two dimensions: computational and statistical complexity. While theoretical researchers often prioritize statistical efficiency - minimizing the number of samples needed to reach a desired accuracy - practitioners focus on reducing computational costs, such as training time and resource consumption. Closing this gap requires algorithms that balance both aspects effectively. In this paper, we introduce MetaStep, a meta-algorithm designed to enhance state-of-the-art RL algorithms by improving their computational performance while maintaining competitive sample efficiency. MetaStep is based on the novel notion of $W$-step Markov decision process (MDP), where, instead of performing a single action and transitioning to the next state, the agent executes a sequence of $W$ actions before observing the resulting state and collecting the discounted $W$-step cumulative reward. First, we provide a theoretical analysis of the suboptimality introduced in the optimal policy performance when planning in a $W$-step MDP, highlighting the impact of the environment stochasticity. Second, we apply MetaStep to GPOMDP, a well-known policy gradient method, and theoretically investigate the advantages of learning in the $W$-step MDP in terms of variance reduction and improved sample complexity. Finally, empirical evaluations confirm that MetaStep reduces computational costs while preserving - and, in certain scenarios, improving - sample efficiency.
URL: https://openreview.net/forum?id=9ynBcLjt7p
---
Title: ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning
Abstract: We present ThinkPrune, a simple yet effective method for pruning the thinking length for long-thinking LLMs, which has been found to often produce inefficient and redundant thinking processes. Existing preliminary explorations of reducing thinking length primarily focus on forcing the thinking process to early exit, rather than adapting the LLM to optimize and consolidate the thinking process, and therefore the length-performance tradeoff observed so far is sub-optimal. To fill this gap, ThinkPrune offers a simple solution that continuously trains the long-thinking LLMs via reinforcement learning (RL) with an added token limit, beyond which any unfinished thoughts and answers will be discarded, resulting in a zero reward. To further preserve model performance, we introduce an iterative length pruning approach, where multiple rounds of RL are conducted, each with an increasingly more stringent token limit. We observed that ThinkPrune results in a remarkable performance-length tradeoff on the AIME24 dataset, the reasoning length of DeepSeek-R1-Distill-Qwen-1.5B can be reduced by half with only 2% drop in performance. We also observed that after pruning, the LLMs can bypass unnecessary steps while keeping the core reasoning process complete.
URL: https://openreview.net/forum?id=V51gPu1uQD
---
Title: TempFlex: Advancing MLLMs with Temporal Perception and Natively Scalable Resolution Encoding
Abstract: Multimodal large language models (MLLMs) have made significant progress across vision-language tasks, yet many designs still suffer from two core limitations. (i) Excessive visual tokens and broken global context: Tiled Patch Encoding fragments high-resolution images, leading to token overload and disrupting global attention modeling. (ii) Lack of temporal reasoning: Most models process video as independent frames using static image encoders, failing to capture temporal dynamics. We present TempFlex-VL, a token-efficient and temporally aware MLLM that addresses both issues through lightweight architectural enhancements. First, we introduce a resolution-agnostic visual encoder that directly processes full images without tiling, preserving global context while substantially reducing visual tokens. Second, we propose Temporal Fiber Fusion (TFF), a plug-and-play module with three complementary pathways: (1) a dynamic local-convolution branch for fine-grained motion, (2) a gated memory accumulator for long-term dependencies, and (3) a periodic encoder for modeling cyclic patterns. These signals are softly fused, enabling the model to adapt to diverse temporal structures without overfitting. To support large-scale video-language pretraining, we curate TempFlex-2M, a high-quality synthetic video–text corpus generated in a single stage via GPT-4o with direct visual prompting. We instantiate TempFlex-VL using two different language backbones, Gemma3-4B and Qwen3-4B, demonstrating the generality of our design across architectures. Both variants achieve state-of-the-art or competitive results on a wide range of image and video benchmarks while markedly improving token efficiency. We will release all code, models, and data to spur future research in unified multimodal understanding.
URL: https://openreview.net/forum?id=ietYdtRB3h
---
Title: FedDUAL: A Dual-Strategy with Adaptive Loss and Dy- namic Aggregation for Mitigating Data Heterogeneity in Federated Learning
Abstract: Federated Learning (FL) marks a transformative approach to distributed model training by combining locally optimized models from various clients into a unified global model. While FL preserves data privacy by eliminating centralized storage, it encounters significant challenges such as performance degradation, slower convergence, and reduced robustness of the global model due to the heterogeneity in client data distributions. Among the various forms of data heterogeneity, label skew emerges as a particularly formidable and prevalent issue, especially in domains such as image classification. To address these challenges, we begin with comprehensive experiments to pinpoint the underlying issues in the FL training process. Based on our findings, we then introduce an innovative dual-strategy approach designed to effectively resolve these issues. First, we introduce an adaptive loss function for client-side training, meticulously crafted to preserve previously acquired knowledge while maintaining an optimal equilibrium between local optimization and global model coherence. Secondly, we develop a dynamic aggregation strategy for aggregating client models at the server. This approach adapts to each client's unique learning patterns, effectively addressing the challenges of diverse data across the network. Our comprehensive evaluation, conducted across three diverse real-world datasets, coupled with theoretical convergence guarantees, demonstrates the superior efficacy of our method compared to several established state-of-the-art approaches.
URL: https://openreview.net/forum?id=8M3XfmNhTZ
---
Title: Augmented Vision-Language Models: A Systematic Review
Abstract: Recent advances in visual-language machine learning models have demonstrated exceptional ability to use natural language and understand visual scenes by training on large, unstructured datasets. However, this training paradigm cannot produce interpretable explanations for its outputs, requires retraining to integrate new information, is highly resource-intensive, and struggles with certain forms of logical reasoning. One promising solution involves integrating neural networks with external symbolic information systems, forming neural symbolic systems that can enhance reasoning and memory abilities. These neural symbolic systems provide more interpretable explanations to their outputs and the capacity to assimilate new information without extensive retraining. Utilizing powerful pre-trained Vision-Language Models (VLMs) as the core neural component, augmented by external systems, offers a pragmatic approach to realizing the benefits of neural-symbolic integration. This systematic literature review aims to categorize techniques through which visual-language understanding can be improved by interacting with external symbolic information systems.
URL: https://openreview.net/forum?id=DFnPi77v6J
---
Title: Adaptive Mesh Quantization for Neural PDE Solvers
Abstract: Physical systems commonly exhibit spatially varying complexity, presenting a significant challenge for neural PDE solvers. In traditional numerical methods, adaptive mesh refinement addresses this challenge by increasing node density in dynamic regions, thereby allocating more computational resources where needed. However, for graph neural operators, this is not always a feasible or optimal strategy. We therefore introduce a novel approach to this issue: rather than modifying grid resolution, we maintain a fixed mesh while dynamically adjusting the bit-width used by a quantized model. We propose an adaptive bit-width allocation strategy driven by a lightweight auxiliary model that identifies high-loss regions in the input mesh. This enables dynamic resource distribution in the main model, where regions of higher difficulty are allocated increased bit-width, optimizing computational resource utilization. We demonstrate our framework's effectiveness by integrating it with two state-of-the-art models, MP-PDE and GraphViT, to evaluate performance across multiple tasks: 2D Darcy flow, large-scale unsteady fluid dynamics in 2D, steady-state Navier–Stokes simulations in 3D, and a 2D hyper-elasticity problem.
Our framework demonstrates consistent Pareto improvements over uniformly quantized baselines, yielding up to 50\% improvements in performance at the same cost.
URL: https://openreview.net/forum?id=NN17y897WG
---
Title: DynFed: Dynamic Test-Time Adaptation for Federated Learning with Adaptive Rate Networks
Abstract: Test-Time Personalized Federated Learning (TTPFL) has emerged as a promising approach for adapting models to distribution shifts in federated learning (FL) environments without relying on labeled data during testing. However, existing methods often struggle with heterogeneous shifts across clients and lack the flexibility to handle diverse distribution changes effectively. In this paper, we introduce DynFed, a novel algorithm that dynamically optimizes test-time adaptation (TTA) in FL scenarios with heterogeneous distribution shifts. Our method leverages Adaptive Rate Networks (ARNs) to generate client-specific adaptation rates, enabling more effective handling of diverse shift types, including label skew and feature shifts. DynFed employs an innovative iterative adaptation process, where adaptation rates are continuously refined based on the current adaptation state using the ARN function, without direct access to raw client data. Crucially, we uncover a fundamental dichotomy: optimal adaptation strategies for one-type and multi-type distribution shifts can be diametrically opposed. DynFed navigates this challenge by automatically adjusting its approach based on the nature of the encountered shifts. Extensive experiments demonstrate that DynFed significantly outperforms existing TTPFL and TTA methods across various shift scenarios. Our theoretical analysis provides convergence and generalization guarantees for our approach and justifies the need
for adaptive mechanisms. Our method shows particularly robust performance in complex multi-type shift environments, where previous approaches often struggle. This work opens new avenues for adaptive and resilient FL in real-world applications where distribution shifts are diverse and unpredictable.
URL: https://openreview.net/forum?id=Np8Jy9kf1b
---
Title: Advancing Event Forecasting through Massive Training of Large Language Models: Challenges, Solutions, and Broader Impacts
Abstract: Many recent papers have studied the development of superforecaster-level event forecasting LLMs. While methodological problems with early studies cast doubt on the use of LLMs for event forecasting, recent studies with improved evaluation methods have shown that state-of-the-art LLMs are gradually reaching superforecaster-level performance, and reinforcement learning has also been reported to improve future forecasting. Additionally, the unprecedented success of recent reasoning models and Deep Research-style models suggests that technology capable of greatly improving forecasting performance has been developed. Therefore, based on these positive recent trends, we argue that the time is ripe for research on large-scale training of superforecaster-level event forecasting LLMs.
We discuss two key research directions: training methods and data acquisition. For training, we first introduce three difficulties of LLM-based event forecasting training: noisiness-sparsity, knowledge cut-off, and simple reward structure problems. Then, we present related ideas to mitigate these problems: hypothetical event Bayesian networks, utilizing poorly-recalled and counterfactual events, and auxiliary reward signals. For data, we propose aggressive use of market, public, and crawling datasets to enable large-scale training and evaluation. Finally, we explain how these technical advances could enable AI to provide predictive intelligence to society in broader areas. This position paper presents promising specific paths and considerations for getting closer to superforecaster-level AI technology, aiming to call for researchers' interest in these directions.
URL: https://openreview.net/forum?id=OeAyoB47dS
---
Title: Tukey g-and-h neural network regression for non-Gaussian data
Abstract: This paper addresses non-Gaussian regression with neural networks via the use of the Tukey g-and-h distribution. The Tukey g-and-h transform is a flexible parametric transform with two parameters $g$ and $h$ which, when applied to a standard normal random variable, introduces both skewness and kurtosis, resulting in a distribution commonly called the Tukey g-and-h distribution. Specific values of $g$ and $h$ produce good approximations to other families of distributions, such as the Cauchy and student-t distributions. The flexibility of the Tukey g-and-h distribution has driven its popularity in the statistical community, in applied sciences and finance. In this work we consider the training of a neural network to predict the parameters of a Tukey g-and-h distribution in a regression framework via the minimization of the corresponding negative log-likelihood, despite the latter having no closed-form expression. We demonstrate the efficiency of our procedure in simulated examples and apply our method to a real-world dataset of global crop yield for several types of crops. Finally, we show how we can carry out a goodness-of-fit analysis between the predicted distributions and the test data. A Pytorch implementation is made available on Github and as a Pypi package.
URL: https://openreview.net/forum?id=sjDmzLoKvj
---
Title: MS-IMAP - A Multi-Scale Graph Embedding Approach for Interpretable Manifold Learning
Abstract: Deriving meaningful representations from complex, high-dimensional data in unsupervised settings is crucial across diverse machine learning applications. This paper introduces a framework for multi-scale graph network embedding based on spectral graph wavelets that employs a contrastive learning approach. We theoretically show that in Paley-Wiener spaces on combinatorial graphs, the spectral graph wavelets operator provides greater flexibility and control over smoothness compared to the Laplacian operator, motivating our approach. A key advantage of the proposed embedding is its ability to establish a correspondence between the embedding and input feature spaces, enabling the derivation of feature importance. We validate the effectiveness of our graph embedding framework on multiple public datasets across various downstream tasks, including clustering and unsupervised feature importance.
URL: https://openreview.net/forum?id=pc6BgWrCjp
---
Title: carps: A Framework for Comparing N Hyperparameter Optimizers on M Benchmarks
Abstract: Hyperparameter Optimization (HPO) is crucial to develop well-performing machine learning
models. In order to ease prototyping and benchmarking of HPO methods, we propose carps,
a benchmark framework for Comprehensive Automated Research Performance Studies
allowing to evaluate N optimizers on M benchmark tasks. In this first release of carps,
we focus on the four most important types of HPO task types: blackbox, multi-fidelity,
multi-objective and multi-fidelity-multi-objective. With 3 336 tasks from 5 community
benchmark collections and 28 variants of 9 optimizer families, we offer the biggest go-to
library to date to evaluate and compare HPO methods. The carps framework relies on a
purpose-built, lightweight interface, gluing together optimizers and benchmark tasks. It
also features an analysis pipeline, facilitating the evaluation of optimizers on benchmarks.
However, navigating a huge number of tasks while developing and comparing methods can
be computationally infeasible. To address this, we obtain a subset of representative tasks by
minimizing the star discrepancy of the subset, in the space spanned by the full set. As a
result, we propose an initial subset of 10 to 30 diverse tasks for each task type, and include
functionality to re-compute subsets as more benchmarks become available, enabling efficient
evaluations. We also establish a first set of baseline results on these tasks as a measure for
future comparisons. With carps (https://anonymous.4open.science/r/CARP-S-860C),
we make an important step in the standardization of HPO evaluation.
URL: https://openreview.net/forum?id=AuA8m4I6zI
---
Title: A Survey on Federated Fine-Tuning of Large Language Models
Abstract: Large Language Models (LLMs) have demonstrated impressive success across various tasks. Integrating LLMs with Federated Learning (FL), a paradigm known as FedLLM, offers a promising avenue for collaborative model adaptation while preserving data privacy. This survey provides a systematic and comprehensive review of FedLLM. We begin by tracing the historical development of both LLMs and FL, summarizing relevant prior research to set the context. Subsequently, we delve into an in-depth analysis of the fundamental challenges inherent in deploying FedLLM. Addressing these challenges often requires efficient adaptation strategies; therefore, we conduct an extensive examination of existing Parameter-Efficient Fine-tuning (PEFT) methods and explore their applicability within the FL framework. To rigorously evaluate the performance of FedLLM, we undertake a thorough review of existing fine-tuning datasets and evaluation benchmarks. Furthermore, we discuss FedLLM's diverse real-world applications across multiple domains. Finally, we identify critical open challenges and outline promising research directions to foster future advancements in FedLLM. This survey aims to serve as a foundational resource for researchers and practitioners, offering valuable insights into the rapidly evolving landscape of federated fine-tuning for LLMs. It also establishes a roadmap for future innovations in privacy-preserving AI. We actively maintain a GitHub repo to track cutting-edge advancements in this field.
URL: https://openreview.net/forum?id=rnCqbuIWnn
---
Title: Learning to Prompt Your Domain for Federated Vision-Language Models
Abstract: The prompt tuning paradigm, with its great advantages of low parameter count and stable training, has recently inspired numerous applications of CLIP-like vision-language models in federated learning. However, in this work, we posit that under significant domain gaps across federated participants, prompt-based CLIP may easily collapse to non-optimal solutions due to the neglect of domain-aware knowledge. We present a novel prompt tuning method, termed ADAPT, to address this issue by learning both intra- and inter-domain prompts. Specifically, we assign each federated participant a domain-specific prompt and use the image's visual features as a condition to guide the generation of language features, with the underlying idea that the prompted CLIP should detect the input image's domain correspondence before making the prediction of its category. Extensive experiments demonstrate ADAPT's significant efficiency and effectiveness in federated learning. For example, by learning and sharing only 2.1M parameters, ADAPT attains a 69.8% average accuracy over the six domains of DomainNet, which improves the original CLIP accuracy by 16.2%.
URL: https://openreview.net/forum?id=OS7zPOZjr3
---