Modeling Retinal Ganglion Cells with Neural Differential Equations
This work addresses efficient neural modeling for vision prosthetics, though it is incremental as it applies existing methods to a new domain.
The paper tackled modeling retinal ganglion cell activity using Liquid Time-Constant Networks and Closed-form Continuous-time Networks, achieving lower MAE, faster convergence, and smaller model sizes compared to convolutional and LSTM baselines across three datasets.
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.