RASP: Read-All, Sparse-Plasticity Continual Learning in a Stationary-Size LoRA
Challenges the one-LoRA-branch-per-session premise with a single stationary-size adapter.
4 items · 3 sources · 3 days
Operational story trace
Follow in this browser to see new updates on your Live feed.
Latest change
Nov 18: two papers move continual learning toward deployment, one adding specialist vision models to a trained multimodal agent's tool space, the other (Hypnos) running continual learning end to end on edge devices.
October's papers framed continual learning around forgetting, replay selection and agent benchmarks. November turned to cheaper adapters: RASP (Nov 2) keeps a single fixed-size LoRA instead of one branch per session, and a Nov 9 paper shares one prompt across tasks with no rehearsal data.
Arc
Challenges the one-LoRA-branch-per-session premise with a single stationary-size adapter.
Drops rehearsal data entirely, anchoring behavior through one shared prompt.
Brings the problem to agents: adding specialist vision models to an already trained agent's tools.
Moves from methods to a platform, targeting continual learning on edge devices.
Challenges the one-LoRA-branch-per-session premise with a single stationary-size adapter.
Drops rehearsal data entirely, anchoring behavior through one shared prompt.
Brings the problem to agents: adding specialist vision models to an already trained agent's tools.
Moves from methods to a platform, targeting continual learning on edge devices.
What to watch — open questions
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.