EFC++: Elastic Feature Consolidation with Prototype Re-balancing for Cold Start Exemplar-free Incremental Learning

Name: EFC++: Elastic Feature Consolidation with Prototype Re-balancing for Cold Start Exemplar-free Incremental Learning
Uploaded: 2026-10-03T13:27:40.523889+00:00
Duration: 333 s
Description: EFC++ tackles Cold Start in exemplar-free incremental learning by regularizing feature drift using an Empirical Feature Matrix (EFM). It consolidates feature representations by penalizing movement along directions critical for past tasks while using prototypes to reduce task-recency bias. This approach decouples backbone training from classifier learning, improving stability and plasticity.

EFC++ tackles Cold Start in exemplar-free incremental learning by regularizing feature drift using an Empirical Feature Matrix (EFM). It consolidates feature representations by penalizing movement along directions critical for past tasks while using prototypes to reduce task-recency bias. This approach decouples backbone training from classifier learning, improving stability and plasticity.

Open the full episode