Modeling and tractable computation form two fundamental but competing pillars of data science; indeed, fitting models to data is often computationally challenging in modern applications. At the same time, a "good" model is one that imposes the right kind of structure on the underlying data-generating process, and this involves trading off multiple competing objectives, e.g., interpretability with flexibility. With a focus on balancing such tensions, I present tractable methodological solutions for fitting flexible models in some canonical machine learning tasks.
The bulk of the talk will focus on a class of “permutation-based" models, which present a flexible alternative to parametric modeling in a host of inference problems involving data generated by people. I introduce a set of algorithmic tools that handles structured missing data and breaks a conjectured computational barrier, demonstrating that carefully chosen non-parametric structure can significantly improve robustness to mis-specification while maintaining interpretability. To conclude the talk, I draw on this perspective to study modeling and computation in high-dimensional regression and reinforcement learning. A focus on exploiting structure in these contexts draws attention to both statistical and computational trade-offs.
Ashwin Pananjady is a PhD student in the Department of Electrical Engineering and Computer Sciences at the University of California Berkeley, advised by Martin Wainwright and Thomas Courtade. His interests lie broadly in statistics, machine learning, information theory, and optimization, and include ranking and permutation estimation, high-dimensional statistics, and reinforcement learning. He is a recipient of the inaugural Lawrence Brown PhD student award from the Institute of Mathematical Statistics, an Outstanding Graduate Student Instructor award from UC Berkeley, and the Governor's Gold Medal from IIT Madras.
Faculty Host: David Forsyth