English

Training and Generating Neural Networks in Compressed Weight Space

Machine Learning 2022-01-03 v1 Computation and Language

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.

Keywords

Cite

@article{arxiv.2112.15545,
  title  = {Training and Generating Neural Networks in Compressed Weight Space},
  author = {Kazuki Irie and Jürgen Schmidhuber},
  journal= {arXiv preprint arXiv:2112.15545},
  year   = {2022}
}

Comments

Presented at ICLR 2021 Workshop on Neural Compression, https://openreview.net/forum?id=qU1EUxdVd_D

R2 v1 2026-06-24T08:36:59.159Z