Harvard Machine Learning Foundations Group

We are a research group focused on some of the foundational questions in modern machine learning. We are interested in both experimental and theoretical approaches that advance our understanding. Our group contains ML practitioners, theoretical computer scientists, statisticians, and neuroscientists, all sharing the goal of placing machine and natural learning on firmer foundations, and elucidating their fundamental capabilities and limitations.

We also run a research-level seminar series on recent advances in the field. Our seminar is affiliated with and supported by the Kempner Institute for the study of natural and artificial intelligence. Join the seminar mailing list for talk announcements.

Opportunities: We are looking for graduate students and postdocs. See opportunities section below. Announcements on positions will also be posted on social media.

People

Researchers

Avatar

Gustaf Ahdritz

PhD Student

Avatar

Alex Atanasov

PhD Student

Avatar

Demba Ba

Faculty

Avatar

Boaz Barak

Faculty

Avatar

Blake Bordelon

PhD Student

Avatar

Chi-Ning Chou

PhD Student

Avatar

Ben Edelman

PhD Student

Avatar

Gal Kaplun

PhD Student

Avatar

Anat Kleiman

PhD Student

Avatar

Depen Morwani

PhD Student

Avatar

Yonadav Shavit

PhD Student

Avatar

Nikhil Vyas

Postdoctoral Fellow

Avatar

Sheng Yang

Masters Student

Avatar

Rosie Zhao

PhD Student

Affiliated

Avatar

Na Li

Faculty

Emeritus

Avatar

Preetum Nakkiran

PhD Student

Avatar

Yamini Bansal

PhD Student

Avatar

Sharon Qian

PhD Student

Avatar

Tristan Yang

Undergraduate

Avatar

Fred Zhang

PhD Student

Coming Soon

Avatar

Sitan Chen

Faculty

Recent Publications

By our group and its members.

(This list is not comprehensive. Also, we’re sometimes slow in updates—see individual homepages and the arXiv for the latest publications.)

Hidden Progress in Deep Learning: SGD Learns Parities Near the Computational Limit

Contrasting random and learned features in deep Bayesian linear regression

Deconstructing Distributions: A Pointwise Framework of Learning

Depth induces scale-averaging in overparameterized linear Bayesian neural networks

Neural Networks as Kernel Learners: The Silent Alignment Effect

Inductive Biases and Variable Creation in Self-Attention Mechanisms

Capacity of Group-invariant Linear Readouts from Equivariant Representations: How Many Objects can be Linearly Classified Under All Possible Views?

Revisiting Model Stitching to Compare Neural Representations

Learning Curves for SGD on Structured Features

Out-of-Distribution Generalization in Kernel Regression

Asymptotics of Representation Learning in Finite Bayesian Neural Networks

For Self-supervised Learning, Rationality Implies Generalization, Provably

The Deep Bootstrap: Good Online Learners are Good Offline Generalizers

Distributional Generalization: A New Kind of Generalization

Learning From Strategic Agents: Accuracy, Improvement, and Causality

Deep Double Descent: Where Bigger Models and More Data Hurt

SGD on Neural Networks Learns Functions of Increasing Complexity

More Data Can Hurt for Linear Regression: Sample-wise Double Descent

Computational Limitations in Robust Classification and Win-Win Results

Minnorm training: an algorithm for training over-parameterized deep neural networks

Adversarial Robustness May Be at Odds With Simplicity

On the Information Bottleneck Theory of Deep Learning

Recent & Upcoming Talks

TBD

TBD

TBD

TBD

TBD

TBD

TBD

TBD

TBD

Deep learning algorithms are responsible for a technological revolution in a variety of tasks, yet understanding why they work remains …

The brain is a very noisy place: when a spike arrives at a pre-synaptic terminal, about half the time neurotransmitter fails to …

The emergence of machines that seem to offer the same or better capabilities than humans raised interests in many sectors who are eager …

Network data is ubiquitous, and many examples can be found in domains ranging from biology to social sciences. Learning from graph data …

This talk gives an overview of distribution-free predictive inference, using conformal prediction. Conformal prediction essentially …

Language models can be dramatically improved by reward models, which predict the quality of a sample. Two approaches for combining …

Seminar Calendar

Below is the calendar of events in the Kempner ML Foundations seminar. Join the mailing list for talk announcements.

Opportunities

We are looking for undergraduate researchers, graduate students and postdocs in the ML foundations group.

For undergraduate students, we are only able to work with students at Harvard or MIT (with preference to the former). If you are a Harvard or MIT student interested in collaborating, informally or formally, with us, please fill out the following google form. Students might also be interested in taking Boaz’s Spring 2023 seminar on the foundations of deep learning.

For graduate students we have openings in Computer Science, Electrical Engineering,applied mathematics or statistics degrees. New: Kempner Institute Graduate Fellowship: See more details here

If you are applying for graduate studies in CS and are interested in machine learning foundations, please mark both “Machine Learning” and “Theory of Computation” as areas of interest. Please also list the names of faculty you want to work with on your application. ML foundations group faculty include Demba Ba (Electrical Engineering and Bioengineering), David Alvarez-Melis, Boaz Barak, Sitan Chen, Jonathan Frankle, Sham Kakade (Computer Science), Cengiz Pehlevan (Applied Mathematics), and Lucas Janson (Statistics). There are also ML foundations affiliated faculty in all of the above departments and more. All of us are also open to the possibilities of co-advising students, including across different departments and schools.

The faculty search in the 2022-3 academic year (for positions starting in Fall 2023) in the school of engineering and applied sciences is focused on climate and energy, which includes also more topics related to machine learning such as energy-system modeling, sustainable algorithms, and more. See here for more details

Postdoc opportunities for 2023-2024 Academic year:

There are a number of opportunities at Harvard for postdoc positions. Applying to multiple positions is not just allowed but encouraged, and we urge you to apply to any of those that are of interest to you.

Follow us on social media for announcements of more opportunities.