raminmh/liquid_time_constant_networks

Implements multiple continuous-time recurrent architectures—LTCs, Neural ODEs, CT-RNNs, and continuous-time GRUs—trainable via backpropagation through time using TensorFlow 1.14. Includes benchmarking scripts across five temporal sequence datasets (gesture recognition, occupancy detection, activity recognition, traffic forecasting, ozone prediction) with standardized evaluation pipelines that output detailed metrics including best-epoch checkpoints and test performance comparisons.

1,812 stars. No commits in the last 6 months.