Postdoc Position in Machine Learning – Qatar Computing Research Institute
Abdelkader Baggag
We are currently looking for a postdoc to join our multi-disciplinary team of researchers and engineers working in Machine Learning at the Qatar Computing Research Institute (QCRI), part of Hamad Bin Khalifa University (HBKU) under Qatar Foundation. We are seeking candidates with expertise in deep learning, generative modeling (GANs, AE, VAE), domain adaptation, optimal transport, and multi-modal modeling.
Project description. Broadly speaking, the goal of our project is to develop optimal transport machinery for time series data, targeting applications in healthcare analytics. This ambitious project aims to accomplish dual objectives of (1) transitioning optimal transport from mathematical theory to computational practice in statistical applications and (2) demonstrating practical value in healthcare analytics and other applications whose structural properties are similar (e.g., computational biology, particle physics, and computer vision). Our target applications have many properties in common, principally the presence of highly stochastic, time-varying data that consists not just of a single point but rather of clouds of information describing the system state or environment. Key milestones include:
o Efficient algorithms for computing transport distances and derived quantities in the statistical setting.
o Theoretical and empirical analysis of the proposed algorithms on dynamic data collected from QCRI’s health team and related sources.
o Extension to combinatorial domains like graphs.
o Incorporation of entropic regularization and other techniques for stability and efficiency.
In the particular case of large scale health data where we expect distributions to not differ much from one another between most pairs of times, there is the possibility of developing new fast and scalable algorithms that use previously constructed transport maps as a bootstrap, as well as GPU-optimized implementations of these algorithms. These algorithms for optimal transport will demonstrate how time series analysis---the key technical application in this proposal---is naturally suited for efficient OT beyond the general case.
Recent methods for learning with a Wasserstein loss are robust to distribution noise; incorporating these methods may increase the reliability of our methods applied to health data and include uncertainty estimates critical for health-related decision-making. When the noise model is unknown or inexact, we can also experiment with cycle-consistency constraints applied to transport maps, a recent method appearing in the geometry and deep learning communities involving analysis of a full dataset rather than individual pairs of points. We expect the resulting maps to be significantly more robust to noise, and it is amenable to smoothing operations typical in time series analysis while incorporating pairwise state similarity.
Analytics Tool. As a final product, we are hoping to build an analytics tool that applies state-of-the-art OT techniques to time-series data expressed as sequences of states. This tool will reflect our newly developed statistical learning techniques and accompanying algorithms. This analytics tool will be used in the context of the ongoing QCRI health project.
Job Summary. The successful candidate will be researching, designing, and developing novel computational models, algorithms, and innovative applications to process data to solve real-world problems, in the health domain. The successful candidate is expected to demonstrate a commitment to deliver results, adaptability, and the ability to work in a teaming environment.
Requirements:
- PhD in computer science, machine learning, computer engineering, or a related field.
- Strong knowledge of deep learning, generative modeling (GANs, AE, VAE), domain adaptation, optimal transport, etc.
- Good knowledge of explainable machine learning.
- Strong record of major accomplishments and publications in top-tier venues.
- Must be highly proficient in Python, PyTorch, TensorFlow.
- Ability to think creatively, work collaboratively, and take ownership of projects.
- Excellent spoken and written English skills and good presentation/communication skills.
Package & Tenure. QCRI offers a unique opportunity for strong research careers and a highly competitive compensation package including attractive tax-free salary and additional benefits such as private health insurance with global coverage, annual paid leave (37 days of annual leave plus 12 days of religious/national holidays), and more for a specified term of 2 years.
Deadline. The position will be open until filled.
How to Apply. Please provide enough information relevant to the requirements of this position to allow the assessment panel to determine your suitability. Send your CV and cover letter to:
Dr. Abdelkader Baggag -contact <aba...@hbku.edu.qa>