Back to search
Top Contributors
247
commits
Compositional Replay Using Memory Blocks (CRUMB) is a replay algorithm for online stream learning that stores and replays training examples at a feature level using a differentiable codebook to mitigate catastrophic forgetting. This repo also contains instructions for setting up the Toybox and iLab-2M-Light stream learning benchmarks.
Stars
2
Forks
0
Watchers
2
Open Issues
0
Overall repository health assessment
No package.json found
This might not be a Node.js project
247
commits
Update README with new dataset benchmarks and ablation experiments
1abd28aView on GitHubscript for easier pretraining without direct loss (ablation study component)
627add9View on GitHub