Using control theory to say which part of a state space model's recurrent state carries a behaviour: directions ranked by how strongly the past writes them and the future reads them, tested by removing a direction at every step.
Data Center Behind-the-meter Energy Hub dispatch problem formulated as two interacting world models, one for the gas-turbine fleet and one for the AI data hall, planned jointly through a differentiable rollout for MPC and RL control.
Using control theory to make language-model components predictable enough for safety-critical deployment: safety classifiers that can prove their decisions, and fine-tuning adapters whose memory can be analyzed and compressed with guarantees.
AI agents turn colocation SLA contracts into verifiable compliance rules, and a per-customer attention model watches power and temperature data to flag likely breaches 30 minutes before they happen.
Safety guarantees for autonomous drones and air taxis: learned reachable sets that build collision avoidance into the controller, and camera-based 3D models from neural radiance fields for safe path planning.
An imitation-learning system that learns how expert operators make decisions in Oil & Gas SCADA operations from multi-sensor data, so those decisions can be made autonomously.
My doctoral research at Purdue: fast, data-driven ways to prove that learned, networked and multi-agent autonomous systems stay safe, and to expose how attackers could exploit them.
Estimators that keep tracking accurate when sensor data is lost in transit, including losses that cluster in predictable regions, and that let a network of sensors agree on one estimate. Demonstrated on aircraft tracking.
No projects match this filter.