Deconstructing Liquid Neural Networks (LNNs)

Explore continuous-time differential equations, Liquid Time-Constant models, pure PyTorch implementations, and chaotic time-series benchmarks.

Neural ODEs & LTCs

Moving from discrete recurrent networks to the continuous mathematics of bounded differential models and Liquid Time-Constants.

Approximating continuous ODEs with discrete numeral solvers to build a functional Liquid layer in pure PyTorch.
View Code on GitHub

Evaluating parameter extreme efficiency by pitting custom LNNs against traditional LSTMs on noisy, chaotic trajectory tasks.