Daily TMLR digest for Jun 05, 2024
TMLR
New certifications
==================
Featured Certification: Linear Bandits with Memory
Giulia Clerici, Pierre Laforgue, Nicolò Cesa-Bianchi
https://openreview.net/forum?id=CrpDwMFgxr
---
Accepted papers
===============
Title: Linear Bandits with Memory
Authors: Giulia Clerici, Pierre Laforgue, Nicolò Cesa-Bianchi
Abstract: Nonstationary phenomena, such as satiation effects in recommendations, have mostly been modeled using bandits with finitely many arms. However, the richer action space provided by linear bandits is often preferred in practice. In this work, we introduce a novel nonstationary linear bandit model, where current rewards are influenced by the learner's past actions in a fixed-size window. Our model, which recovers stationary linear bandits as a special case, leverages two parameters: the window size $m \ge 0$, and an exponent $\gamma$ that captures the rotting ($\gamma < 0)$ or rising ($\gamma > 0$) nature of the phenomenon. When both $m$ and $\gamma$ are known, we propose and analyze a variant of OFUL which minimizes regret against cyclic policies. By choosing the cycle length so as to trade-off approximation and estimation errors, we then prove a bound of order $\sqrt{d}\,(m+1)^{\frac{1}{2}+\max\{\gamma,0\}}\,T^{3/4}$ (ignoring log factors) on the regret against the optimal sequence of actions, where $T$ is the horizon and $d$ is the dimension of the linear action space. Through a bandit model selection approach, our results are then extended to the case where both $m$ and $\gamma$ are unknown. Finally, we complement our theoretical results with experiments comparing our approach to natural baselines.
URL: https://openreview.net/forum?id=CrpDwMFgxr
---
Title: Simple Imputation Rules for Prediction with Missing Data: Theoretical Guarantees vs. Empirical Performance
Authors: Dimitris Bertsimas, Arthur Delarue, Jean Pauphilet
Abstract: Missing data is a common issue in real-world datasets. This paper studies the performance of impute-then-regress pipelines by contrasting theoretical and empirical evidence. We establish the asymptotic consistency of such pipelines for a broad family of imputation methods. While common sense suggests that a 'good' imputation method produces datasets that are plausible, we show, on the contrary, that, as far as prediction is concerned, crude can be good. Among others, we find that mode-impute is asymptotically sub-optimal, while mean-impute is asymptotically optimal. We then exhaustively assess the validity of these theoretical conclusions on a large corpus of synthetic, semi-real, and real datasets. While the empirical evidence we collect mostly supports our theoretical findings, it also highlights gaps between theory and practice and opportunities for future research, regarding the relevance of the MAR assumption, the complex interdependency between the imputation and regression tasks, and the need for realistic synthetic data generation models.
URL: https://openreview.net/forum?id=IKH5ziX9dk
---
Title: DIGNet: Learning Decomposed Patterns in Representation Balancing for Treatment Effect Estimation
Authors: Yiyan HUANG, WANG Siyi, Cheuk Hang LEUNG, Qi WU, Dongdong WANG, Zhixiang Huang
Abstract: Estimating treatment effects from observational data is often subject to a covariate shift problem incurred by selection bias. Recent research has sought to mitigate this problem by leveraging representation balancing methods that aim to extract balancing patterns from observational data and utilize them for outcome prediction. The underlying theoretical rationale is that minimizing the unobserved counterfactual error can be achieved through two principles: (I) reducing the risk associated with predicting factual outcomes and (II) mitigating the distributional discrepancy between the treated and controlled samples. However, an inherent trade-off between the two principles can lead to a potential loss of information useful for factual outcome predictions and, consequently, deteriorating treatment effect estimations. In this paper, we propose a novel representation balancing model, DIGNet, for treatment effect estimation. DIGNet incorporates two key components, PDIG and PPBR, which effectively mitigate the trade-off problem by improving one aforementioned principle without sacrificing the other. Specifically, PDIG captures more effective balancing patterns (Principle II) without affecting factual outcome predictions (Principle I), while PPBR enhances factual outcome prediction (Principle I) without affecting the learning of balancing patterns (Principle II). The ablation studies verify the effectiveness of PDIG and PPBR in improving treatment effect estimation, and experimental results on benchmark datasets demonstrate the superior performance of our DIGNet model compared to baseline models.
URL: https://openreview.net/forum?id=Z20FInfWlm
---
Title: Mildly Overparameterized ReLU Networks Have a Favorable Loss Landscape
Authors: Kedar Karhadkar, Michael Murray, Hanna Tseran, Guido Montufar
Abstract: We study the loss landscape of both shallow and deep, mildly overparameterized ReLU neural networks on a generic finite input dataset for the squared error loss. We show both by count and volume that most activation patterns correspond to parameter regions with no bad local minima. Furthermore, for one-dimensional input data, we show most activation regions realizable by the network contain a high dimensional set of global minima and no bad local minima. We experimentally confirm these results by finding a phase transition from most regions having full rank Jacobian to many regions having deficient rank depending on the amount of overparameterization.
URL: https://openreview.net/forum?id=10WARaIwFn
---
Title: Augmenting Ad-Hoc IR Dataset for Interactive Conversational Search
Authors: Pierre ERBACHER, Jian-Yun Nie, Philippe Preux, Laure Soulier
Abstract: A peculiarity of conversational search systems is that they involve mixed-initiatives such as system-generated query clarifying questions. Evaluating those systems at a large scale on the end task of IR is very challenging, requiring adequate datasets containing such interactions. However, current datasets only focus on either traditional ad-hoc IR tasks or query clarification tasks, the latter being usually seen as a reformulation task from the initial query.
Only few datasets are known to contain both document relevance judgments and the associated clarification interactions such as Qulac and ClariQ. Both are based on the TREC Web Track 2009-12 collection but cover a very limited number of topics
(237 topics), far from being enough for training and testing conversational IR models.
To fill the gap, we propose a methodology to automatically build large-scale conversational IR datasets from ad-hoc IR datasets in order to facilitate explorations on conversational IR.
Our methodology is based on two processes: 1) generating query clarification interactions through query clarification and answer generators, and 2) augmenting ad-hoc IR datasets with simulated interactions.
In this paper, we focus on MsMarco and augment it with query clarification and answer simulations. We perform a thorough evaluation showing the quality and the relevance of the generated interactions for each initial query. This paper shows the feasibility and utility of augmenting ad-hoc IR datasets for conversational IR.
URL: https://openreview.net/forum?id=z8d7nT1HWw
---
Title: Text Descriptions are Compressive and Invariant Representations for Visual Learning
Authors: Zhili Feng, Anna Bair, J Zico Kolter
Abstract: Modern image classification is based on directly predicting classes via large discriminative networks, which do not directly contain information about the intuitive visual features that may constitute a classification decision. Recently, work in vision-language models (VLM) such as CLIP has provided ways to specify natural language descriptions of image classes, but typically focuses on providing single descriptions for each class. In this work, we demonstrate that an alternative approach, in line with humans' understanding of multiple visual features per class, can also provide compelling performance in the robust few-shot learning setting. In particular, we introduce a novel method, \textit{SLR-AVD (Sparse Logistic Regression using Augmented Visual Descriptors)}. This method first automatically generates multiple visual descriptions of each class via a large language model (LLM), then uses a VLM to translate these descriptions to a set of visual feature embeddings of each image, and finally uses sparse logistic regression to select a relevant subset of these features to classify each image. Core to our approach is the fact that, information-theoretically, these descriptive features are more invariant to domain shift than traditional image embeddings, even though the VLM training process is not explicitly designed for invariant representation learning. These invariant descriptive features also compose a better input compression scheme. When combined with finetuning, we show that SLR-AVD is able to outperform existing state-of-the-art finetuning approaches in both in-distribution and out-of-distribution tasks.
URL: https://openreview.net/forum?id=spo705Fyv0
---
New submissions
===============
Title: Preconditioned Neural Posterior Estimation for Likelihood-free Inference
Abstract: Simulation based inference (SBI) methods enable the estimation of posterior distributions when the likelihood function is intractable, but where model simulation is feasible. Popular neural approaches to SBI are the neural posterior estimator (NPE) and its sequential version (SNPE). These methods can outperform statistical SBI approaches such as approximate Bayesian computation (ABC), particularly for relatively small numbers of model simulations. However, we show in this paper that the NPE methods are not guaranteed to be highly accurate, even on problems with low dimension. In such settings the posterior cannot be accurately trained over the prior predictive space, and even the sequential extension remains sub-optimal. To overcome this, we propose preconditioned NPE (PNPE) and its sequential version (PSNPE), which uses a short run of ABC to effectively eliminate regions of parameter space that produce large discrepancy between simulations and data and allow the posterior emulator to be more accurately trained. We present comprehensive empirical evidence that this melding of neural and statistical SBI methods improves performance over a range of examples including a motivating example involving a complex agent-based models applied to real tumour growth data.
URL: https://openreview.net/forum?id=vgIBAOkIhY
---