Improve Forward Transfer by Selectively Concatenating Prompts from Prior Tasks in Continual Learning
-
Updated
Feb 8, 2024 - Jupyter Notebook
Improve Forward Transfer by Selectively Concatenating Prompts from Prior Tasks in Continual Learning
Implements dynamic and expandable architectures for continual learning — where networks grow or prune over tasks to balance plasticity and stability. Includes Progressive Neural Networks, Dynamically Expandable Networks (DEN), and PackNet, with utilities for parameter freez
To associate your repository with the progressive-networks topic, visit your repo's landing page and select "manage topics."