Certified Behavioral Retention for Continual Adaptation with Held-Out Anchors
Paper · https://arxiv.org/abs/2610.40740
arXiv:2610.40740v1 Announce Type: new Abstract: Continual adaptation requires deciding not only how to update a model, but whether to commit an update without exceeding a declared budget for changes to protected behavior. We introduce Certified Anchor-Based Adaptation (CABA), a candidate-agnostic post-update gate that reserves historical observations from candidate optimization, compares old and candidate policies on frozen representations, and selects the largest tested interpolation coefficient whose worst-anchor forward KL divergence remains within budget. The finite check requires neither action labels nor complete trajectories. We develop a theory-led certification framework that separates measured anchor retention from conditional off-anchor guarantees. Under explicit representation-coverage and Lipschitz assumptions, an accepted update satisfies a uniform behavioral-divergence bound on protected support. The analysis gives regularity bounds for affine softmax heads, probabilistic coverage and compression guarantees, interpolation monotonicity conditions, grid-resolution bounds, and trajectory-level and sequential-composition guarantees. In a controlled UCI Human Activity Recognition study used as a non-stationary sensor-policy proxy, CABA reduces mean forgetting from 1.01 percent for unconstrained adaptation to 0.00 percent at budget epsilon = 0.05, while reducing final average accuracy from 93.04 percent to 88.48 percent. The experiment illustrates a retention-plasticity trade-off and does not establish foundation-model or robotic performance. CABA provides an auditable retention check for candidate generators exposing comparable policy distributions and a declared update path, while deployment also requires separate evidence of current-task utility.
First seen by Oriel on .
AI summary
The paper introduces CABA, a post-update gate that certifies continual adaptations against protected-behavior change budgets using held-out anchors and interpolation.
- CABA reserves historical observations from candidate optimization, compares old and candidate policies on frozen representations, and accepts the largest tested interpolation coefficient within a forward-KL budget.
- Certification requires neither action labels nor complete trajectories, making it applicable to candidate generators that expose comparable policy distributions and a declared update path.
- Under representation-coverage and Lipschitz assumptions, the framework provides off-anchor behavioral-divergence, trajectory-level, sequential-composition, grid-resolution, and interpolation guarantees.
- In a controlled UCI Human Activity Recognition proxy, mean forgetting fell from 1.01 percent to 0.00 percent at epsilon = 0.05, while final average accuracy fell from 93.04 percent to 88.48 percent.
- The experiment demonstrates a retention-plasticity trade-off but does not establish foundation-model or robotic performance; deployment still requires separate evidence of current-task utility.
#continual learning #policy adaptation #behavioral retention #forward kl
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2610.40740v1 | https://arxiv.org/abs/2610.40740 Read discussion |