All Projects
Browse current research across Healthcare AI, World Models, and AI for Science.
Featured Projects
PHAST
A Port-Hamiltonian neural architecture for stable long-horizon forecasting in dynamical systems.
GRL-SNAM
Geometric reinforcement learning for navigation and mapping in unknown environments.
AI4PD: Parkinson's Disease
AI4PD: patient-specific AI for Parkinson's disease. The program hub, with the patient portal, the digital-twin approach, clinician workflow, evidence, and partners.
Current Projects
![]()
A Prior-Robust Sensitivity and Factorial-Necessity Analysis of Tri-Modal Viral-Ultrasound Gene Delivery for Lysosomal Neurodegeneration
A four-part audit of the tri-modal GM2 model: under corrected priors, enzyme kinetics, not BBB entry, dominate, and focused ultrasound is practically negligible by day 365.
![]()
PD Research Companion
Two 2026 manuscripts on Parkinson disease: longitudinal validity gates and signed dopaminergic asymmetry, with outcome-blind patient dossiers.
![]()
MICCAI 2026 Posterior-Aware Motor Phenotyping
Posterior-aware Bayesian motor-state phenotyping from MDS-UPDRS-III with DaTSCAN, MRI, and BioFIND validation.
![]()
QC-PHAST Search
An evidence-gated protocol for comparing classical and quantum query models in finite-pool rare-regime discovery.
![]()
Piano Digital Twin
A performer-instrument world model for multimodal piano skill, structured dynamics, and safe personalized training.
![]()
Ski Jumper Digital Twin
A physics-aware digital twin for phase-dependent ski-jump trajectories, aerodynamic posture, and landing safety.
![]()
A Novel Tri-Modal Viral–Ultrasound Gene-Delivery Therapy Protocol for Lysosomal Neurodegeneration via Stochastic Model Optimization with Uncertainty Quantification and Generalizability
Computationally optimized tri-modal AAV-T4, SP2, and focused ultrasound protocol for lysosomal neurodegeneration with stochastic uncertainty quantification.
![]()
Material-Aware Hamiltonian Risk Fields
A context-energy term that makes a port-Hamiltonian navigation policy activate evasive maneuvers only when the local scene admits a safer feasible route.
![]()
Event-Triggered Hamiltonian Optimization
SHAPE, a port-Hamiltonian optimizer that treats stagnation as a control event to hunt for better minima under a fixed evaluation budget.
![]()
The Physics, Information, and Computation of Perennial Learning: Kolmogorov Complexity, Information Distance and Port-Hamiltonian Thermodynamics
Perennial learning framework using Kolmogorov complexity, information distance, and port-Hamiltonian dynamics for safe continual adaptation.
![]()
Scalable Risk-Averse Well-Placement
Scalable risk-aware optimization for choosing well placements in shale fields.
![]()
Differential and Pointwise Control RL
Sample-efficient reinforcement learning that combines differential control updates with pointwise trajectory corrections.
![]()
Computer Algebra Meets Hamiltonian Geometry
Symbolic computation for deriving, simplifying, and visualizing Hamiltonian and Lie-Poisson dynamical systems.
![]()
Dynamic Metric Space Maintenance
A self-balancing dynamic octree giving logarithmic-time updates and queries in evolving metric spaces, without rebuilding as distributions shift.
![]()
Dynamic Belief Games
Decision-making models for rapid response in dynamic and adversarial communication networks.
![]()
Subsurface Flow Modeling
Computational models for simulating and visualizing subsurface flow in complex geological formations.
![]()
OC Protein Side-Chain and Folding
Continuous reinforcement learning for protein side-chain packing and folding.
![]()
Night-time Aerial Material Segmentation
Night-time aerial material segmentation that combines hyperspectral sensing with RGB imagery.
![]()
Scalable Robust Bayesian Co-Clustering
Variational co-clustering for discovering robust row and column structure in noisy, sparse, and high-dimensional data.
![]()
4DRecons
Complete textured deforming 3D models from a single RGB-D sequence, using 4D neural implicit surfaces with deformation and topology regularizations.
- (opens in a new tab)
![]()
Real Time Processing of Hyperspectral Video
Real-time machine learning for analyzing hyperspectral video streams as they are captured.
![]()
Dueling Neural ODEs
Stable neural ODE methods for learning forward and adjoint dynamics together.
![]()
Theoretical Bound for OCF Algorithm
Theory for understanding when and why optimal control flow algorithms work.
Past Projects
![]()
MCE-VAE
A variational learning framework for discovering stable latent clusters in transformed data.
![]()
Video Imputation
Methods for repairing missing or corrupted video frames during streaming, compression, and super-resolution.
![]()
Physics-informed Neural Networks
Physics-informed neural networks for solving PDEs while respecting governing physical constraints.
![]()
Sample Complexity
Theory and algorithms for reinforcement learning that achieve better performance with fewer samples.
![]()
Optimal Control
Computational methods for solving optimal control problems through Hamiltonian system dynamics.
![]()
DEDRECON
Real-time anomaly detection for hyperspectral video using encoder-decoder models that fuse spectral and motion cues.
![]()
Adversarial Cloaking
Adversarial patches for 3D human meshes that hide people from machine vision systems.
![]()
Camera ISP
Learning-based image signal processing for removing heterogeneous artifacts in camera pipelines.
![]()
Angstrom
Multi-resolution geometric data structures for scientific computing and large-scale spatial analysis.
![]()
Spectre
Fast spectral and geometric processing with data structures and prediction methods that come with guarantees.
![]()
Generative Material Design
Generative models for designing new materials from limited labeled data.
![]()
Rank-ordered Search-and-Score
Multi-agent reinforcement learning for finding important regions in gigapixel images quickly.
![]()
SHASTRA
Collaborative tools for geometric modeling, simulation, visualization, and design on networked systems.
- (opens in a new tab)
![]()
DiDi: Data Intensive Display Intensive Computing
Archived CVC project on data-intensive, display-intensive visualization for very large scientific datasets and multi-tiled display environments.
- (opens in a new tab)
![]()
VisualEyes
Archived CVC project on visual environments, domain modeling, auralization, and interrogative visualization for scientific data.
