Selected Research Projects

Efficient State-Space Models

Efficient State-Space Models

We develop methods for learning compact, high-performing state-space models for sequential data. Our work leverages control-theoretic principles to compress and optimize dynamical systems during training, enabling efficient, expressive models that generalize across tasks such as long-range sequence modeling, speech, text, and audio. In particular, we emphasize techniques that dynamically identify and retain the most influential latent dimensions, accelerating training while preserving task-critical structure.

Project CETI

Project CETI: Autonomous Multi-Agent Wildlife Monitoring

Project CETI is a research endeavor aiming to learn the language of sperm whales and reveal insights into their social behavior. Our team develops autonomous aerial robotic systems to monitor and track whales in the wild, combining decentralized multi-robot coordination, real-time detection, and game-theoretic planning.

Robust Flight Navigation

Robust Flight Navigation Out-of-Distribution

We develop methods for robust vision-based flight navigation that transfer beyond the training environment, even under severe distribution shifts or directly from simulation. Our work leverages liquid neural networks—a class of continuous-time, dynamical models inspired by biological neurons—to distill task-relevant visual features and discard irrelevant information, enabling strong real-world generalization.