README.md

Supporting materials for "Get rich quick: exact solutions reveal how unbalanced initializations promote rapid feature learning"

This repository contains code generating figures in the NeurIPS 2024 Spotlight paper titled "Get rich quick: exact solutions reveal how unbalanced initializations promote rapid feature learning".

You can access the full paper on arXiv.

Citation

If you use this work in your research, please cite the paper as follows:

@misc{kunin2024rich,
      title={Get rich quick: exact solutions reveal how unbalanced initializations promote rapid feature learning}, 
      author={Daniel Kunin and Allan Raventós and Clémentine Dominé and Feng Chen and David Klindt and Andrew Saxe and Surya Ganguli},
      year={2024},
      eprint={2406.06158},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2406.06158}, 
}