Patrick LaVictoire
Mathematician and machine learning leader
Berkeley, California
Summary
Deep quantitative foundation in pure mathematics with a focused research record in ergodic theory and harmonic analysis. escholarship+2
Applied machine learning practitioner who transitioned from academia to industry, working on location prediction, fraud detection, and other production ML systems at Lyft and Apple. lyft+1
Contributor to AI alignment research and agendas during a research fellowship at MIRI, coauthoring survey and technical material connecting ML practice to alignment concerns. intelligence+1
Machine learning leader and manager in industry with roles spanning individual contributor engineering to ML management, emphasizing rigorous modeling and reliable data pipelines. getprog+1
Work
Education
Projects
Writing
Alignment for advanced machine learning systems
January 1, 2016MIRI research report surveying research areas for aligning increasingly autonomous learning systems with operator interests; co-authored as part of MIRI's alignment agenda.
Robust Cooperation in the Prisoner's Dilemma: Program Equilibrium via Provability Logic
January 1, 2014Paper applying modal provability logic to design agents (modal agents) that can robustly achieve mutual cooperation in one-shot Prisoner's Dilemmas between algorithms with access to each other's source code.
Pointwise Ergodic Theorems for Nonconventional L^1 Averages
January 1, 2010Doctoral thesis proving new results on pointwise convergence of nonconventional ergodic averages in L^1, including results on averages along Bernoulli random sequences and demonstrating nonconvergence cases for polynomial and prime sequences.