Research

Each plate runs in three panels: the problem, the machinery, and the result. The objects carry across the panels, so a single mark can be followed the whole way through. Work done inside a role is written up under that role on Experience.

The subject is observed at four irregular visits. The event happens after the last one, at an unknown time.Each observation pins a latent diffusion whose clock is set by the conditions. The event is its first passage across the boundary.Solving the Kolmogorov equations on a grid gives the event-time density exactly, so the reported interval is calibrated rather than nominal.

Neural Stochastic Differential Equations for Produce Quality

First author · in preparation

An avocado is photographed every few days until it spoils. From those few looks, predict the day it turns, with an interval you can plan against. The hidden state is a diffusion; spoilage is the moment it first crosses a line, solved exactly rather than simulated.

neural SDEs · first-passage times · Fokker–Planck equations · interval censoring

Repository

Benchmarks record difficulty in three buckets. Problems inside one bucket are not remotely alike.Item-response theory and Glicko-2 re-estimate every problem on one continuous standardized scale, from human and model attempt records.Accuracy against that scale is a curve, not a scalar. It shows exactly where each model stops generalizing.

Easy2Hard-Bench

Research assistant · NeurIPS 2024, Datasets & Benchmarks

Benchmarks call a problem easy, medium or hard, which tells you almost nothing. This puts difficulty on a continuous scale learned from how people and models actually did, so accuracy becomes a curve and you can see exactly where a model gives out.

item-response theory · Glicko-2 rating · difficulty calibration · generalization

Paper Repository Dataset

Match play is scored by hole, not by stroke. A blow-up costs one hole, and a birdie wins exactly one.Two policies leave the same tee: one solved blind to the opponent, one solved against their shot distribution.Across every match state the gain is near zero. It exists in one narrow band, and opponent statistics that precise are not obtainable.

A Game-Theoretic Study of Adversary Strategy in Golf

First author · Computational Statistics, 2024

Should a professional play differently because of who they are across from? Posed as a two-player game on tour shot data, the answer is almost always no: the edge is real in a narrow band of situations and swamped everywhere else by not knowing your opponent well enough.

game theory · Markov decision processes · stochastic games · value iteration

Paper Preprint Repository

Every arrival is sorted at the door: admitted, or sent home. Six attributes are on the record when that happens.Shapley values weigh what each attribute contributed. Insurance status carried more than the severity index itself.Matched on severity, admission rate still differs by insurance program. Language, against expectation, was not a barrier.

Using Model Classification to Detect Bias in Hospital Triaging

Co-author · Annals of Biomedical Science and Engineering, 2023 · JUST Health, Stanford School of Medicine, 2022

Who gets admitted at the emergency room door, and what actually decided it. At the same severity, patients on federally run insurance were admitted less often than those on private plans. Language, against expectation, was not a barrier.

Shapley attribution · game-theoretic interpretability · ensemble classification

Paper JUST Health Repository

Scholar LinkedIn GitHub