https://github.com/pxl-th/yama.jl

Visualize loss landscape
https://github.com/pxl-th/yama.jl

flux julia loss-landscape neural-networks visualization

Last synced: 18 days ago
JSON representation

Visualize loss landscape

• Host: GitHub
• URL: https://github.com/pxl-th/yama.jl
• Owner: pxl-th
• License: mit
• Created: 2020-06-15T19:57:29.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2020-06-19T13:18:02.000Z (about 6 years ago)
• Last Synced: 2025-02-27T02:51:14.710Z (over 1 year ago)
• Topics: flux, julia, loss-landscape, neural-networks, visualization
• Language: Julia
• Homepage:
• Size: 922 KB
• Stars: 1
• Watchers: 3
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Yama 山

[![Build Status](https://github.com/pxl-th/Yama.jl/workflows/CI/badge.svg)](https://github.com/pxl-th/Yama.jl/actions)

[![](https://img.shields.io/badge/docs-latest-blue.svg)](https://pxl-th.github.io/Yama.jl/dev/)

Visualize loss surface in random directions (1st & 2nd picture) or between model checkpoints (3rd picture).



  

  



Visualize path that optimizer took during training (4th picture).



  

  



## References

- [Hao Li et.al. "Visualizing the Loss Landscape of Neural Nets"](https://arxiv.org/abs/1712.09913)

- [Original code in Python](https://github.com/tomgoldstein/loss-landscape)