Body & Behavior Twins

Intelligent behavior ultimately unfolds through bodies moving in the physical world. We extend the digital-twin paradigm beyond neural circuits to motor behavior and physical systems: learning probabilistic 3D pose estimators from multi-view video, reconstructing biomechanical trajectories with calibrated uncertainty, and modeling behavioral idiosyncrasies in both humans and animals. In parallel, we combine deep learning with physics-based forward models — using finite-element simulations for electrical impedance tomography (EIT) of the lung — to build hybrid models that respect known physical constraints while learning complex structure from data.