CoLLAs 2026 keynote: fix forgetting via drift compensation and model merging, not just prevention
apsarathchandar · x · 2026-09-15
In the second Day 1 keynote at CoLLAs 2026 in Bucharest, Joost van de Weijer argued for a fresh take on continual learning: instead of only fighting catastrophic forgetting, use drift compensation to correct representations after the fact. He also shared insights on model merging to combine independently trained task models into one.
More from Research
- Kevin Murphy unifies goal-based hierarchical RL with ACGVF and his 25-year-old HHMM work — sirbayes · 2026-09-15
- New paper: LLM projects to enhance democracy threaten democratic autonomy — sethlazar · 2026-09-15
- MIT team builds recursive meta-intelligence AI that invents its own scientific instruments — ProfBuehlerMIT · 2026-09-15
- Nature Medicine paper: on-premise medical AI agents for reliable clinical decisions — EricTopol · 2026-09-15
- Mathematician predicts two parallel pillars of math: human understanding vs incomprehensible Lean proofs — thegautamkamath · 2026-09-15
- Consciousness giant Antonio Damasio publishes new book Natural Intelligence today — AnnaCiaunica · 2026-09-15