Elastic Weight Consolidation

Selectively slows learning on parameters critical to earlier tasks, reducing catastrophic forgetting via a quadratic regulariser.

---

iCaRL – Incremental Classifier and Representation Learning

Learns new classes incrementally with exemplar rehearsal + nearest-mean classifier; no catastrophic forgetting.

---

Learning without Forgetting

Uses knowledge distillation to preserve performance on old tasks while training on new tasks — no access to old data.

---

← Back to projects