cornn-convex-rnn-optimization

What problem does it solve?

Convex optimization converts non-convex RNN training into a convex formulation, dramatically speeding up training on standard hardware.

Core Features & Use Cases

• Convexified RNN training enables rapid inference of neural dynamics.
• Supports million-parameter RNNs on standard hardware.
• Real-time network reconstruction for large-scale neural data.
• Use cases include large-scale neural recordings, neural dynamics inference, and attractor structure recovery.

Quick Start

Provide neural_data and call cornn_train with chosen hidden_dim and regularization to train the CORNN model and inspect the resulting W_rec, W_in and dynamics.