This repository contains basic experimental and prelimiary codes for various ML concepts and ideas, including neural network merging, hyper networks and meta learning, neural network connectivity . This is essentially a playground for me to explore the ML concepts and test some of isntant ideas to study an currently ongoing project: "generation of knowledge".
Clone the repository via git as follows:
git clone https://github.com/minkiml/Knowledge-generation.git
Then, install requirements as follows:
pip install -r requirements.txt
The current implementation uses a single GPU and provides no multiple GPU setup.
All runs are made on common CV datasets, including MNIST, CIFAR-10, CIFAR-100, SVHN, STL10, and CelebA.
Download the datasets from their original source and specify the data directory to each run script (.sh) file.
See each folder for detailed readme.