hiteshvaidya/feed-forward-network-for-continual-learning

This repository contains codebase for a basic Feed Forward Network in tensorflow without using keras that operates in a continual learning setting

Feed-Forward-Network-for-continual-learning

This repository contains codebase for a basic Feed Forward Network that operates in a continual learning setting. The code is developed from scratch without using keras library. This can also be adapted to build a vanilla feed forward neural network that is trained in a normal i.i.d. setting.

Please refer to this blog for a step-by-step guide.

Installation

Clone this repository

git clone [email protected]:hiteshvaidya/Feed-Forward-Network-for-continual-learning.git

Create a new environment with python 3.10

conda create -n lifelong python=3.10
conda activate lifelong

Install all the necessary libraries

pip install -r requirements.txt

Structure

lifelong
├── data
│   ├── cifar10
│   ├── fashion
│   ├── iid
│   ├── kmnist
│   └── mnist
├── requirements.txt
└── src
    ├── __pycache__
    ├── cifarConv.py
    ├── layer.py
    ├── main.py
    ├── network.py
    └── util.py

Launching experiments

python main.py -d mnist -n 20 -lr 0.01 -b 32