Contrastive Consolidation of Top-Down Modulations Achieves Sparsely Supervised Continual Learning

Viet Anh Khoa Tran, Emre O. Neftci, Willem A. M. Wybo
Neuromorphic Software Ecosystems (PGI-15), Research Center Jülich

[Overview] [arXiV]

This is the PyTorch implementation of the task-modulated contrastive learning (TMCL) algorithm. TMCL provides a novel approach to sparsely supervised continual learning by continually integrating supervised top-down modulations with contrastive learning.

Installation

pip install -r requirements.txt
pip install -e .

Usage

Our run scripts assume a SLURM envionment. If you want to run the code on a local machine, you can use the --local flag in the scripts (not tested).

python slurm/submit_cifar100_s5.py.