https://github.com/pythonnut/hypernet
HyperGAN extension to large target models
https://github.com/pythonnut/hypernet
Last synced: about 1 month ago
JSON representation
HyperGAN extension to large target models
- Host: GitHub
- URL: https://github.com/pythonnut/hypernet
- Owner: PythonNut
- Created: 2019-04-18T18:31:45.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2019-08-24T04:23:49.000Z (over 5 years ago)
- Last Synced: 2025-02-07T08:48:59.387Z (3 months ago)
- Language: Python
- Homepage:
- Size: 85 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# HyperGAN Extensions
This project was done for Advanced Big Data Analysis, Math 389L at CGU.
We worked on extending [HyperGAN](https://arxiv.org/abs/1901.11058).
In particular, we introduced:
* A dynamic hypernet constructor that can produce hypernets for arbitrary target networks, allowing us to experiment on a much larger ResNet20 target network.
* Training discrete target networks from a minibatch instead of treating them as a minibatch sized ensemble.
* Pretraining for the encoder and generators, to ensure that the initial distribution of weights breaks symmetry in multiple directions.