Physics inspired machine learning
- Discretized ODE-based learning frameworks for modeling sequetial inputs and outputs.
- CoRNN (Coupled Oscillatory Recurrent Neural Networks) and UnicoRNN for learning sequences with very long-term dependencies.
- LEM (Long Expressive Memory) for multi-scale sequential learning.
- Discretized PDE-based Graph Neural Networks.
- Equilibrium propagation (EqProp) for neuromorphic machine learning.
Selected Publications
T. K. Rusch and S. Mishra, Coupled Oscillatory Recurrent Neural Network (coRNN): An accurate and gradient stable architecture for learning long-time dependencies, International Conference on Learning Representations (ICLR), 2021 (Orals). available from arXiv:2010.00951.
T. K. Rusch and S. Mishra, UniCORNN: A recurrent model for learning very long time dependencies, International Conference on Machine Learning (ICML), 2021, available from ArXiv:.2103.05487.
T. K. Rusch, S. Mishra, N. B. Erichson and M. Mahoney, Long Expressive Memory for sequence modeling, Preprint 2021, arXiv:04744.