Projects

Selected research in visual learning, anomaly detection, representation structure, language models, reinforcement learning, and uncertainty quantification.

VLAAD

A lightweight collision-aware plug-in that learns temporally localized risk representations from captioned driving video.

Read project

Diagram of exact group maps being refitted to observed feature transformations

Fitting choices shape the predictive cost of exact group models

Testing whether the predictive penalty of exact group models comes from the group laws—or from how the models are fitted.

Read project

Poster summarizing out-of-distribution experiments on distilled chain-of-thought language models

Improving the out-of-distribution performance of distilled chain-of-thought language models

Testing whether small distilled language models generalize beyond their training distributions—and whether Fishr regularization can help.

Read project

F1 scores for two time-series anomaly detection models across random seeds

Unveiling the Flaws

A critical analysis of how window size, random seeds, and normalization can dominate reported anomaly-detection performance.

Read project

Evaluation returns for Continual-DreamerV3 and two DreamerV3 baselines across alternating tasks

Dream On: Continual reinforcement learning with DreamerV3

Decoupling DreamerV3’s critic into permanent and transient components to balance adaptation with knowledge retention.

Read project

Poster for a project using SVM surrogates for uncertainty quantification

Applying support vector machines to uncertainty quantification

Using SVM regression as a surrogate for Monte Carlo circuit simulation, with substantially fewer training samples.

Read project