Modeling Retinal Ganglion Cells with Neural Differential Equations

TL;DR

This paper introduces neural differential equation models, LTCs and CfCs, for retinal ganglion cell activity, demonstrating improved efficiency and adaptability over traditional models, suitable for edge applications like vision prosthetics.

Contribution

It presents novel neural differential equation architectures for retinal modeling, showing advantages in efficiency and adaptability over existing methods.

Findings

Lower MAE compared to baselines

Faster convergence and smaller models

Effective in limited data and retraining scenarios

Abstract

This work explores Liquid Time-Constant Networks (LTCs) and Closed-form Continuous-time Networks (CfCs) for modeling retinal ganglion cell activity in tiger salamanders across three datasets. Compared to a convolutional baseline and an LSTM, both architectures achieved lower MAE, faster convergence, smaller model sizes, and favorable query times, though with slightly lower Pearson correlation. Their efficiency and adaptability make them well suited for scenarios with limited data and frequent retraining, such as edge deployments in vision prosthetics.