Gradient Diffusion: Enhancing Multicompartmental Neuron Models for Gradient-Based Self-Tuning and Homeostatic Control

Autor:	Landsmeer, Lennart P. L., Negrello, Mario, Hamdioui, Said, Strydis, Christos
Rok vydání:	2024
Předmět:	Quantitative Biology - Neurons and Cognition
Druh dokumentu:	Working Paper
Popis:	Realistic brain models contain many parameters. Traditionally, gradient-free methods are used for estimating these parameters, but gradient-based methods offer many advantages including scalability. However, brain models are tied to existing brain simulators, which do not support gradient calculation. Here we show how to extend -- within the public interface of such simulators -- these neural models to also compute the gradients with respect to their parameters. We demonstrate that the computed gradients can be used to optimize a biophysically realistic multicompartmental neuron model with the gradient-based Adam optimizer. Beyond tuning, gradient-based optimization could lead the way towards dynamics learning and homeostatic control within simulations.
Databáze:	arXiv
Externí odkaz:	http://arxiv.org/abs/2412.07327 Zobrazit plný text záznamu View this record from Arxiv