LIDS & Stats Tea

Tea talks are 20 minute long informal chalk-talks for the purpose of sharing ideas and making others aware about some of the topics that may be of interest to the LIDS and Stats audience. If you are interested in presenting in the upcoming seminars, please email lids_stats_tea[at]mit[dot]edu

February 2, 2022

Datamodels: Understanding Model Predictions as functions of Data

Sung Min (Sam) Park (CSAIL)

Current supervised machine learning models rely on an abundance of training data. Yet, understanding the underlying structure and biases of this data—and how they impact models—remains challenging. We present a new conceptual framework, datamodels,...

February 9, 2022

Near-Optimal Learning of Extensive-Form Games with Imperfect Information

Tiancheng Yu (LIDS)

His paper resolves a longstanding open question of designing near-optimal algorithms for learning imperfect-information extensive-form games from bandit feedback. We present the first line of algorithms that require only $\widetilde{\mathcal{O}}((XA...

February 16, 2022

Symmetric Binary Perceptron Model: Algorithms and Barriers

Eren Can Kizildag (LIDS)

It has been shown very recently that the symmetric binary perceptron (SBP) exhibits an extreme form of clustering at all positive densities: almost all of its solutions are singletons separated by large distances. This suggests that finding a...

February 23, 2022

Intrinsic Dimension Estimation with Optimal Transport

Adam Block (IDSS)

It has long been thought that high-dimensional data encountered in many practical machine learning tasks have low-dimensional structure, i.e., the manifold hypothesis holds. A natural question, thus, is to estimate the intrinsic dimension of a given...

March 2, 2022

Counterfactual inference in sequential experimental design

Raaz Dwivedi (LIDS & Harvard)

We consider the problem of counterfactual inference in sequentially designed experiments wherein a collection of $\mathbf{N}$ units each undergo a sequence of interventions for $\mathbf{T}$ time periods, based on policies that sequentially adapt...

March 9, 2022

WCFS Queues: A New Analysis Framework

Isaac Grosof (CMU)

In this talk, I investigate four queueing models that are key to understanding the behavior of modern computing systems. Each was previously considered intractable to analyze. However, we discovered a subtle similarity between these models, which we...

March 16, 2022

Time varying regression with hidden linear dynamics

Horia Mania (LIDS)

We revisit a model for time-varying linear regression that assumes the unknown parameters evolve according to a linear dynamical system. Counterintuitively, we show that when the underlying dynamics are stable the parameters of this model can be...

March 30, 2022

Non-parametric threshold for smoothed empirical Wasserstein distance

Zeyu Jia (LIDS)

Consider an empirical measure Pn induced by n iid samples from a d-dimensional K-subgaussian distribution P. We show that when K < σ, the Wasserstein distance W2(Pn ∗N(0,σ2Id),P∗N(0,σ2Id)) converges at the parametric rate O(1/n), and when K >...

April 6, 2022

Training invariances and the low-rank phenomenon: beyond linear networks.

Thien Le (LIDS)

The implicit bias induced by the training of neural networks has become a topic of rigorous study. In the limit of gradient flow and gradient descent with appropriate step size, it has been shown that when one trains a deep linear network with...

April 13, 2022

Higher-order network information improves overlapping community detection

Xinyi Wu (IDSS)

Identifying communities is a central problem in network science. While many methods exist, one major challenge in community detection is the often overlapping nature of communities such that nodes belong to multiple groups simultaneously. Link-based...

April 20, 2022

Auditing algorithmic filtering on social media

Sarah Cen (LIDS)

By filtering the content that users see, social media platforms have the ability to influence users' perceptions and decisions, from their dining choices to their voting preferences. This influence has drawn scrutiny, with many calling for...

April 27, 2022

The Climate Pocket: Exploring Local Climate Impacts with Physics-informed Machine Learning

Björn Lütjens (AeroAstro)

Would a global carbon tax reduce the flood risk at MIT? The answer to this question of local impact and risk is critical for policy making or climate-resilient infrastructure development. But, localized climate models are computationally too...

October 26, 2022

A non-asymptotic analysis of oversmoothing in Graph Neural Networks

Xinyi Wu (IDSS & LIDS)

A central challenge of building more powerful Graph Neural Networks (GNNs) is the oversmoothing phenomenon, where increasing the network depth leads to homogeneous node representations and thus worse classification performance. While previous works...

November 2, 2022

On counterfactual inference with unobserved confounding via exponential family

Abhin Swapnil Shah (LIDS)

We are interested in the problem of unit-level counterfactual inference with unobserved confounders owing to the increasing importance of personalized decision-making in many domains: consider a recommender system interacting with a user over time...

November 9, 2022

A Simple and Optimal Policy Design with Safety against Heavy-tailed Risk for Stochastic Bandits

Feng Zhu (IDSS & LIDS)

We design new policies that ensure both worst-case optimality for expected regret and light-tailed risk for regret distribution in the stochastic multi-armed bandit problem. It is recently shown that information-theoretically optimized bandit...

November 16, 2022

The Husky Programming Language for Efficient and Strongly Interpretable AI

Xiyu Zhai (LIDS)

We invent a new programming language called Husky ( for efficient and strongly interpretable AI that can be fundamentally different from deep learning and traditional models. It’s a long term project...

November 30, 2022

Contextual Bandits and Optimistically Universal Learning

Moïse Blanchard (ORC & LIDS)

We study the question of learnability for contextual bandits when the reward function class is unrestricted and provide consistent algorithms for large families of data-generating processes. Our analysis shows that achieving consistency irrespective...

December 7, 2022

Can Direct Latent Model Learning Solve Linear Quadratic Gaussian Control?

Yi Tian (LIDS)

We study the task of learning state representations from potentially high-dimensional observations, with the goal of controlling an unknown partially observable system. We pursue a direct latent model learning approach, where a dynamic model in some...

December 14, 2022

Optimal Learning Rates for Regularized Least-Squares with a Fourier Capacity Condition

Prem Murali Talwai (LIDS & ORC)

We derive minimax adaptive rates for a new, broad class of Tikhonov-regularized learning problems in Hilbert scales under general source conditions. Our analysis does not require the regression function to be contained in the hypothesis class, and...