Technical interest · T-01
Time series
Learning from sequences that unfold in time: market returns, disease counts, ICU stays and vibration signals.
A lot of my modeling work has been on data indexed by time. At MIT CSAIL I trained a financial time-series transformer on daily bars across 499 stocks; at Fort Bend County I forecast influenza, RSV and COVID-19 cases; the ICU risk model is a Markov chain over daily patient states; and the visual-to-tactile work predicts acceleration signals over time.
Nearest neighbors · any layercosine distance in the full space