Publications

My current research often considers problems related to Randomized Numerical Linear Algebra (RandNLA). This area can be regarded as a fusion between more theoretically oriented fields, such as random matrix theory and matrix perturbation theory, with the more practice-oriented field of numerical linear algebra. This area thoroughly captures my goal of studying practical theory. I spend my time learning about mathematically elegant results, such as establishing matrix concentration inequalities via the matrix Laplace transform, and then use the consequences of these results to develop new approaches to ubiquitous computational problems, such as linear optimization.

Prior to starting my PhD, I carried out research in the field of Health Informatics, working with researchers at Regenstrief Institute. While doing so, I asked (and partially answered) questions related to the deployment of machine learning in healthcare. This was an excellent experience, as it emphasized the systems-level challenges (e.g. data privacy, bureaucratic inertia, formatting standards) to effectively using the algorithms we develop to solve end-use problems.

🔸 - Denotes alphabetical author order

Under Review/In Preparation

• mSAM: Micro-Batch-Averaged Sharpness-Aware Minimization
  Under Review
  Kayhan Behdin, Qingquan Song, Aman Gupta, Sathiya Keerthi, Ayan Acharya, Borja Ocejo Elizondo, Gregory Dexter, Rajiv Khanna, David Durfee, and Rahul Mazumder

• Feature Space Sketching for Logistic Regression
  Under Review
  Gregory Dexter, Rajiv Khanna, Jawad Raheel, and Petros Drineas

Published/Accepted

• A Precise Characterization of SGD Stability Using Loss Surface Geometry
  To appear at ICLR 2024
  Gregory Dexter, Borja Ocejo, Sathiya Keerthi, Aman Gupta, Ayan Acharya, Rajiv Khanna

• MaSk-LMM: A Matrix Sketching Framework for Linear Mixed Models in Association Studies
  To appear at RECOMB 2024
  Myson Burch, Aritra Bose, Gregory Dexter, Laxmi Parida, and Petros Drineas

• Sublinear Time Deterministic Algorithms for Spectral Approximation
  ITCS 2024 / To appear in Algorithmica
  🔸 Rajarshi Bhattacharjee, Gregory Dexter, Cameron Musco, Archan Ray, and David Woodruff

• Sketching Algorithms for Sparse Dictionary Learning: PTAS and Turnstile Streaming
  NeurIPS 2023
  🔸 Gregory Dexter, Petros Drineas, David Woodruff, and Taisuke Yasuda

• Sublinear Time Eigenvalue Approximation via Random Sampling
  ICALP 2023
  🔸 Rajarshi Bhattacharjee, Gregory Dexter, Petros Drineas, Cameron Musco, and Archan Ray

• Faster Randomized Interior Point Methods for Tall/Wide Linear Programs
  JMLR
  Agniva Chowhdury, Gregory Dexter, Palma London, Haim Avron, and Petros Drineas

• On the Convergence of Inexact Predictor-Corrector Methods for Linear Programming
  ICML 2022, Selected for long presentation (2% acceptance rate)
  Gregory Dexter, Agniva Chowhdury, Haim Avron, and Petros Drineas

• Inverse Reinforcement Learning in a Continuous State Space with Formal Guarantees
  NeurIPS 2021
  Gregory Dexter, Kevin Bello, and Jean Honorio

• Randomized linear algebra approaches to estimate the von neumann entropy of density matrices
  IEEE Transactions on Information Theory (2020)
  Eugenia-Maria Kontopoulou, Gregory Dexter, Wojciech Szpankowski, Ananth Grama, and Petros Drineas

Health Informatics Research

• Generative Adversarial Networks for Creating Synthetic Free-Text Medical Data: A Proposal for Collaborative Research and Re-use of Machine Learning Models
  IOS 2019 / AMIA 2021
  Suranga Kasthurirathne, Gregory Dexter, and Shaun Grannis

• Generalization of machine learning approaches to identify notifiable conditions from a statewide health information exchange
  Medinfo 2019 / AMIA 2020 (Selected for oral presentation at both venues)
  Gregory Dexter, Shaun Grannis, Brian Dixon, and Suranga Kasthurirathne

• Comparison of Free-Text Synthetic Data Produced by Three Generative Adversarial Networks for Collaborative Health Data Analytics
  AMIA 2019
  Gregory Dexter, Shaun Grannis, and Suranga Kasthurirathne