What kinds of models can be implemented in Jaxley?

Jaxley focuses on biophysical, Hodgkin-Huxley-type models. You can think of Jaxley like the NEURON simulator written in JAX.

Jaxley allows to simulate the following types of models, as well as networks thereof:

• single-compartment (point neuron) Hodgkin-Huxley models

• multi-compartment Hodgkin-Huxley models

• rate-based neuron models (tutorial here)

For all of these models, Jaxley is flexible and accurate. For example, it can flexibly add new channel models, use different kinds of synapses (conductance-based, tanh, …), it can insert different kinds of channels in different branches (or compartments) within single cells, and it can simulate complex ion dynamics (diffusion, pumps,…). Like NEURON, Jaxley implements a backward-Euler solver for stable numerical solution of multi-compartment neurons.

In addition to these biophysical neuron models, Jaxley can also simulate simplified models, see the tutorial here. In particular, Jaxley supports:

• Leaky-integrate-and-fire (LIF) neurons,

• Izhikevich neurons,

• Rate-based neurons.

Jaxley also supports networks of these neurons.

Note that, for LIF and Izhikevich neuron models, Jaxley does not yet support surrogate gradient descent, which is required for efficient training due to the discontinuity of spikes in these models.