Only Strict Saddles in the Energy Landscape of Predictive Coding Networks?

• Code for the NeurIPS 2024 paper Only Strict Saddles in the Energy Landscape of Predictive Coding Networks?.
• See this blog post for the key ideas of the paper.
• Also check out this library for training predictive coding networks, which was used for some of the experiments in this paper.

💻 Reproducibility

All the results and plots from the paper can be reproduced from the included notebooks. All notebooks except for the convergence experiments can be run in < 15min on a CPU. For the heavier convergence experiments, we recommend using a GPU.

• Figure 1:
• For the toy examples in Figure 2 (and the statistics in Figure 7): &
• Figures 3-4 as well as 8-10:
• Figure 5 as well as 11-12:
• Figure 6:

📄 Citation

If you found this code useful in your work, please cite the paper:

@article{innocenti2025only,
  title={Only Strict Saddles in the Energy Landscape of Predictive Coding Networks?},
  author={Innocenti, Francesco and Achour, El Mehdi and Singh, Ryan and Buckley, Christopher L},
  journal={Advances in Neural Information Processing Systems},
  volume={37},
  pages={53649--53683},
  year={2025}
}