For many applications in healthcare, econometrics, financial forecasting and climate science, data can only be obtained as aggregates. This begs the question, can one construct accurate models using only aggregates? I will present two vignettes outlining recent work towards an answer. First, consider a sparse linear model learned from IID …
Distributed implementations of popular machine learning algorithms exhibit poor scaling when deployed on more than a few tens of compute nodes. The key sources of this poor performance are communication bottlenecks and straggler nodes in the system. In this talk, I will explain why these bottlenecks are a real challenge …
Given the dramatic successes in machine learning and reinforcement learning over the past half decade, there has been a resurgence of interest in applying these techniques to continuous control problems in robotics, self-driving cars, and unmanned aerial vehicles. Though such applications appear to be straightforward generalizations of standard reinforcement learning, …
In this talk, I will present ongoing work on the knockoff filter for inference in regression. In a high-dimensional model selection problem, we would like to select relevant features without too many false positives. The knockoff filter provides a tool for model selection by creating knockoff copies of each feature, …
Multi-stage stochastic optimization can be used to model dynamic decision-making environments in which a sequence of decisions are to be made in response to a sequence of random events. Such problems arise in many applications, such as unit commitment and economic dispatch in power systems and inventory and production management. …
This talk considers matrix completion in cases where columns are points on a nonlinear algebraic variety (in contrast to the commonplace linear subspace model). A special case arises when the columns come from a union of subspaces, a model that has numerous practical applications. We propose a new approach to …
Video: https://vimeo.com/201151747 From recommender systems to healthcare analytics low-rank recovery from partial observations is prevalent in modern data analysis. There has been significant progress over the last decade in providing rigorous guarantees for low-rank recovery problems based on convex relaxation techniques. However, the computational complexity of these algorithms render them …
Video: https://vimeo.com/202266237 We present ongoing work regarding best arm identification in multi-armed bandit problems, when the reward distributions are bounded. Although this is a standard assumption in this context, state of the art methods (such as the lil-UCB algorithm) use sub-Gaussian concentration bounds for the mean rewards. However, one can …
Video: https://vimeo.com/203310049 Speaker 1: Urvashi Oswal Title: Network-based whole-brain representational similarity learning Abstract: Technologies such as functional magnetic resonance imaging (fMRI) provide huge amounts of data that could help improve our understanding of the human brain but they are often plagued by many complications, including noise, high-dimensionality, strong and unknown …
Video: https://vimeo.com/204561429 Abstract: In uncertainty propagation, coefficients, boundary conditions, and other key inputs of computational models are given as random variables and one is interested in estimating statistical moments of the corresponding model outputs. Estimating the moments with crude Monte Carlo can become prohibitively expensive in cases where a single …