Training and Generating Neural Networks in Compressed Weight Space

Kazuki Irie, Jürgen Schmidhuber

Keywords: discrete cosine transform, recurrent neural networks, fast weights, language modelling

TL;DR: We revisit RNNs with DCT-encoded weight matrices and its fast weight extension for language modelling.

Abstract:

The inputs and/or outputs of some neural nets are weight matrices of other neural nets. Indirect encodings or end-to-end compression of weight matrices could help to scale such approaches. Our goal is to open a discussion on this topic, starting with recurrent neural networks for character-level language modelling whose weight matrices are encoded by the discrete cosine transform. Our fast weight version thereof uses a recurrent neural network to parameterise the compressed weights.
We present experimental results on the enwik8 dataset.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2112.15545/code)