tomgoldstein/loss-landscape
PyTorch-based tool for visualizing loss surfaces of neural networks near optimal parameters.
Not currently ranked — collecting fresh signals.
star history
This repository provides code for visualizing the loss landscape of neural networks. Given a pre-trained model and its parameters, it calculates and plots loss surfaces along random directions, enabling analysis of network training dynamics. The tool supports distributed computation across multiple GPUs and nodes, storing results in HDF5 format. The work accompanied a NIPS 2018 paper on loss landscape visualization.
Frequently asked
- What is tomgoldstein/loss-landscape?
- PyTorch-based tool for visualizing loss surfaces of neural networks near optimal parameters.
- Is loss-landscape open source?
- Yes — tomgoldstein/loss-landscape is open source, released under the MIT license.
- What language is loss-landscape written in?
- tomgoldstein/loss-landscape is primarily written in Python.
- How popular is loss-landscape?
- tomgoldstein/loss-landscape has 3.2k stars on GitHub.
- Where can I find loss-landscape?
- tomgoldstein/loss-landscape is on GitHub at https://github.com/tomgoldstein/loss-landscape.