Deep learning via dynamics.
We build neuroscience-grounded and geometry-aware learning systems for dynamical signals— from chaotic flows to physiological signals.
Our work draws inspiration from cortical organization, thalamocortical gating, and anatomically constrained computation, while leveraging hyperbolic geometry, topological structure, and operator-theoretic perspectives for stable representation learning.
Selected external ideas we admire—especially those that treat time, neural dynamics, and biological structure as first-class design principles for modern learning systems.
Selected works spanning equivariant learning, neuro-inspired reservoirs, geometry-aware modeling and topological memory. For the complete list, see Publications (full).
- Adaptive Group-Weighted Convolutions: Enhancing Equivariant Networks via Learnable Symmetry Importance Maps
- Brain-Lateralized Reservoirs: Neuro-Anatomically Constrained Echo State Networks for Chaotic and Physiological Signals
- Geometry-Aware Reservoirs: Patch-Wise Jacobian Lifting with Cross-Patch Couplings for Piecewise-Linear Modeling of Chaotic Flows
- Hodge-Projected Echo-State Networks with Topologically Anchored Memory for Chaotic Flows
- Thalamocortical Burst Gating Enables Event-Locked Memory
- Phase-Modulated Permutation Reservoirs for Robust Biomedical Time-Series Forecasting
- Hierarchical Fractal Reservoirs: Unsupervised Multiscale Symbolic Graphs for Long-Horizon Prediction of Chaotic Flows
- Symbolic Ordinal Partition Reservoirs for Model-Free Forecasting of Chaotic Flows
- A Biologically Grounded Reservoir with Provable Dynamics and Slot-Based Working Memory
- HypER: Hyperbolic Echo State Networks for Capturing Stretch-and-Fold Dynamics in Chaotic Flows
- Frozen in Time: Parameter-Efficient Time Series Transformers via Reservoir-Induced Feature Expansion and Fixed Random Dynamics
Active courses and lecture material.
-
Deep Learning and Generative AI CS 922
-
Discrete Mathematics MS 202
-
Machine Learning CS 601
-
Mathematical Foundations for Machine Learning (MFML) Workshop MFML 2026
Curated links to documents, publication lists, and academic materials.
Collaborations are welcome—especially on neuro-inspired learning, geometric representations, dynamical systems, and reservoir computing.