Research · 2026
Systems that notice when they are wrong.
Two research programmes across scientific machine learning and cyber-physical intelligence, documented with their evidence, limits, and reversals intact.
02 research systems34 public log entries
01
VARA-PINN · Submitted
Variable-Aware Regional Adaptive Control for Physics-Informed Neural Networks
A guarded feedback controller that diagnoses where and which physical channel is failing before making a bounded intervention around ordinary PINN training.
02
SafeGraph-CAV · Submitted
Trust-Weighted and Runtime-Shielded CACC for Cyber-Resilient Vehicle Platooning
A four-layer cyber-physical architecture that treats each V2V message as a directed claim, estimates relational trust, uses it inside cooperative control, and supervises the final command with a physical gap/TTC shield.