Topic: Jaxley

Biophysical neuron models provide mechanistic insight about empirically observed phenomena. However, biophysical neuron models are expensive to simulate, thereby limiting the scale of biophysical models, and optimizing their parameters is notoriously difficult, thereby preventing the fitting of biophysical models to physiologically meaningful tasks or datasets (panel a). We built a new toolbox for simulation and inference in neuroscience, Jaxley, which allows to parallelize simulations on (multiple) GPUs and to compute the gradient of simulations with respect to parameters (panel b, c). We use

Advancements in machine learning have drastically improved our ability to fit biophysical models of neural systems. We present here a fully-differentiable, biophysical model of the mouse retina that is primed for gradient-based optimization. Our model is based on the Jaxley software library (Deistler et al. 2024, Bernstein Conference) and JAX machine learning framework. The model consists of a network of Hodgkin-Huxley style neurons, realistic synapse dynamics, an experimentally-informed organization of cells in 3D space, and spatially constrained connectivity. We further implemented a wide v