brainevent 0.0.7

Enabling Event-driven Computation in CPU/GPU/TPU.

Enabling Event-driven Computation in CPU/GPU/TPU

Brain is characterized by the discrete spiking events, which are the fundamental units of computation in the brain.

BrainEvent provides a set of data structures and algorithms for such event-driven computation on CPUs, GPUs, TPUs, and maybe more, which can be used to model the brain dynamics in an efficient and biologically plausible way.

Particularly, it provides the following class to represent binary events in the brain:

• BinaryArray: representing array with a vector/matrix of events.

Furthermore, it implements the following commonly used data structures for event-driven computation of the above class:

• COO: a sparse matrix in COO format for sparse and event-driven computation.
• CSR: a sparse matrix in CSR format for sparse and event-driven computation.
• CSC: a sparse matrix in CSC format for sparse and event-driven computation.
• JITCHomoR: a just-in-time connectivity matrix with homogenous weight for sparse and event-driven computation.
• JITCNormalR: a just-in-time connectivity matrix with normal distribution weight for sparse and event-driven computation.
• JITCUniformR: a just-in-time connectivity matrix with uniform distribution weight for sparse and event-driven computation.
• FixedPreNumConn: a fixed number of pre-synaptic connections for sparse and event-driven computation.
• FixedPostNumConn: a fixed number of post-synaptic connections for sparse and event-driven computation.
• ...

BrainEvent is fully compatible with physical units and unit-aware computations provided in BrainUnit.

Usage

If you want to take advantage of event-driven computations, you must warp your data with brainevent.BinaryArray:

import brainevent

# wrap your array with BinaryArray
event_array = brainevent.BinaryArray(your_array)

Then, the matrix multiplication with the following data structures, $\mathrm{event\ array} @ \mathrm{data}$, will take advantage of event-driven computations:

• Sparse data structures provided by brainevent, like:
  • brainevent.CSR
  • brainevent.JITCHomoR
  • brainevent.FixedPostNumConn
  • ...
• Dense data structures provided by JAX/NumPy, like:
  • jax.numpy.ndarray
  • numpy.ndarray

data = jax.random.rand(...)  # normal dense array
data = brainevent.CSR(...)  # CSR structure
data = brainevent.JITCScalarR(...)  # JIT connectivity
data = brainevent.FixedPostNumConn(...)  # fixed number of post-synaptic connections

# event-driven matrix multiplication
r = event_array @ data
r = data @ event_array

Installation

You can install brainevent via pip:

pip install brainevent -U

Alternatively, you can install BrainX, which bundles brainevent with other compatible packages for a comprehensive brain modeling ecosystem:

pip install BrainX -U

Documentation

The official documentation is hosted on Read the Docs: https://brainevent.readthedocs.io/

See also the ecosystem

brainevent is one part of our brain modeling ecosystem: https://brainmodeling.readthedocs.io/