Some research themes and applied projects that I work on.
Building privacy preserving, auditable and distributed ledger to enable confidential transaction and tokens for instant settlement without a trusted setup.
My phd thesis is on designing multi-camera tracking systems that are able to allocate their own resources intelligently to achieve high tracking performance while reasoning about dynamic scene and resource constraints.
We construct robust policies with k-of-n counterfactual regret minimization subroutine that exhibit non-obvious cautious behavior without task specific safety designs.
We propose explicit shaping and bandit based shaping methods that lets an agent learn from a arbitrary human advice without getting distracted from the original task and accelerating its learning from good advice.
Efficient planning methods for sequential decision-making problems where agents must act under partial information. Emphasis on principled planning, belief-state representations, and performance guarantees.
Additional applied machine learning projects spanning short-horizon market dynamics, sports analytics, and climate-related modeling, typically exploratory and interdisciplinary in nature.
