Code for NeurIPS 2024 Paper:
Abhiram Iyer, Sarthak Chandra, Sugandha Sharma, Ila Fiete. Flexible mapping of abstract domains by grid cells via self-supervised extraction and projection of generalized velocity signals. NeurIPS, 2024.
Instructions
runner.py allows the user to run various experiments. Each experiment can be parameterized through a "config", which is contained in configs.py. The analysis scripts (e.g., analysis_bird2d.ipynb) provide code for generating paper figures and running analyses within each "abstract domain".
Cite our work:
@inproceedings{iyerflexible,
title={Flexible mapping of abstract domains by grid cells via self-supervised extraction and projection of generalized velocity signals},
author={Iyer, Abhiram and Chandra, Sarthak and Sharma, Sugandha and Fiete, Ila R},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems}
}
Wandb runs: https://wandb.ai/iyer/gcpc_velocity/overview