DEFT Introduces Core-Slack Architecture to Prevent Catastrophic Forgetting

DEFT is a new method for continual learning that prevents catastrophic forgetting by partitioning networks into a frozen Core and a trainable Slack using structural masking. By merging redundant neurons and restricting cross-layer interactions, it effectively preserves old knowledge while allowing new learning.

2026-07-24 ~ 2026-07-24 · 3 related posts

• DEFT extends its Core–Slack split across layers to keep transfer learning stable — TheGradient · 2026-07-24
• DEFT splits networks into frozen Core and trainable Slack to block forgetting — TheGradient · 2026-07-24
• DEFT proposes a structural way to prevent catastrophic forgetting in continual learning — TheGradient · 2026-07-24