clane9/sparse-transcoder

Visualizing high-dimensional data with sparse transcoding

Sparse Transcoder

Work in progress, feedback and contributions welcome!

Visualizing high-dimensional data is hard. But then, natural images are high-dimensional, and we can easily understand those? Here we implement sparse transcoding, a method to translate arbitrary high-dimensional data into natural images so we can see the structure in them better.

The basic steps of our method are simple:

1. Train a sparse auto-encoder for the target data (the "encoder").
2. Train a sparse auto-encoder for natural image patches (the "decoder").
3. Visualize by encoding with the encoder and decoding with the decoder ("transcoding").

Installation

Clone the repo and install the package in a new virtual environment

git clone https://github.com/clane9/sparse-transcoder.git
cd sparse-transcoder

python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -r requirements.txt
pip install --no-deps -e .

Inspiration

• Sparse coding
• Sparse autoencoders for interpretability
• Multi-layer sparse modeling

Contribute

If you'd like to contribute, please feel free fork the repo and start a conversation in our issues.