Overview | Quickstart | Installation | Examples | More SITH
SITHshow is a library showcasing applications of the neurally inspired SITH representation of working memory for use in neural networks. Because SITH has a fuzzy memory of the past, it can often outperform RNNs and deep learning models with fixed buffers.
Jump right in by trying the example notebooks in Google Colaboratory!
To install this package with the JAX CPU backend (Windows users without WSL 2), run
pip install --upgrade "jax[cpu]"
git clone https://github.com/compmem/SITHshow
cd SITHshow
pip install .To install this package with the JAX GPU backend (NVIDIA only), run one of the corresponding code chunks compatible with your GPU.
# CUDA 12 installation
# Note: wheels only available on linux.
pip install --upgrade "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
git clone https://github.com/compmem/SITHshow
cd SITHshow
pip install .# CUDA 11 installation
# Note: wheels only available on linux.
pip install --upgrade "jax[cuda11_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
git clone https://github.com/compmem/SITHshow
cd SITHshow
pip install .Requires Python 3.9+ and JAX 0.4.13+. See the JAX Docs for more information about cross-platform compatibility.
Getting started with SITH
Train on one scale, generalize to many with SITHCon
- Morse Decoder (suitable to run on cpu)
- AudioMNIST (more computationally intensive, GPU recommended)
Efficiently learn latent number lines with CNL
- Odometer (suitable to run on cpu)