BrainTrace documentation#

BrainTrace implements scalable online learning for recurrent neural networks (RNNs) and spiking neural networks (SNNs) using eligibility trace propagation (ETP). Choose an entry point below to begin.


Installation#

Choose the platform that will run JAX. BrainTrace requires Python 3.11 or newer.

pip install -U "braintrace[cpu]"

CUDA 12:

pip install -U "braintrace[cuda12]"

CUDA 13:

pip install -U "braintrace[cuda13]"

The NVIDIA driver, CUDA version, and JAX wheel must be compatible.

pip install -U "braintrace[tpu]"
pip install -r requirements.txt

For verification and platform troubleshooting, see Installation.


Learn more#

Choose the entry point that matches your next task.

Installation
Installation
Quickstart
Quickstart
Core Concepts
Core Concepts
RNN Online Learning
RNN Online Learning
SNN Online Learning
SNN Online Learning
Advanced Topics
Compiler Internals

Learning path#

Follow the central path from top to bottom. Branches mark the RNN/SNN workflow and the two algorithm tutorials.

Numbered dependency map of the BrainTrace documentation modules

A dependency guide, not a required linear syllabus.

Foundation

1 Quickstart · 2 Core Concepts

Online Learning Routines

3 RNN Online Learning · 4 SNN Online Learning

Online Learning Algorithms

5 Choose an Algorithm · 6 D-RTRL · 7 pp-prop

Internals and application

8 Operators for Online Learning · 9 RNN Compiler Walkthrough · 10 Examples and Advanced Topics


See also the BrainX ecosystem#

BrainTrace is part of BrainX, an open, differentiable ecosystem for full-scale brain simulation.