You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
An interpretable foundation model of the patient clinical timeline: event forecasting, calibrated time-to-event alerts, and concept-level interpretability, benchmarked head-to-head against tuned GBMs, tabular foundation models, and survival baselines on MIMIC-IV, eICU, and GEMINI via the MEDS standard.
[CVPRW2024 FGVC11 (Best paper award)] Official pytorch implementation of the paper: "ConceptHash: Interpretable Fine-Grained Hashing via Concept Discovery"
Interpretability by construction: route a layer's computation through a certified Legible Bottleneck and emit a runtime Faithfulness Certificate that bounds everything the named concepts cannot explain. Paper and reference implementation.
ISIC 2018 dermoscopy deep learning — all three tasks, a concept-bottleneck experiment, a skin-tone fairness audit, and a scored entry on the live ISIC MILK10k leaderboard (macro F1 0.422). My first DL project, retrained in 2026 with an honest retrospective.
Interpretable spatial graph framework integrating pathway and ligand–receptor priors with tissue architecture. Generates pathway maps and H&E overlays that reveal how tumors organize and rewire signaling in space.
Code and checkpoints for "The Illusion of Control: Why Bare Classifier Inversion Silently Fails in Concept-Bottleneck Text Generation" (EMNLP 2026). Multi-attribute controllable text generation, concept bottleneck, compositional generalization.
Interpretable perturb-seq modeling with a pathway/TF concept bottleneck — predicts effects and identifies stable regulatory drivers that generalize across datasets.