Xun Sun, Thanh Thi Nguyen, Thuy-Trang Vu · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.31401 · cited by 0 works (OpenAlex) · open-access PDF
Непрерывное обучение требует, чтобы модель продолжала приобретать новые навыки и знания, не компрометируя свои существующие возможности. Параметрические методы страдают от катастрофического забвения и потери пластичности. Непараметрические методы, такие как позывы, навыки и приспособления, используют контекстное обучение, но ограничиваются длиной контекста и убыточным сжатием простого текста. Чтобы эффективно сочетать свои дополнительные сильные стороны, мы предлагаем HERD (Harness-Evolving Recursive Destillation), гибридную рамку непрерывного обучения, которая эффективно учится в приспособлении и интернализирует опыт в накопительные возможности путем дистилляции. На стороне приспособления HERD постоянно совершенствует свои навыки для захвата повторных моделей,
Machine translation (Meta M2M100); the English original follows.
Continual learning requires a model to keep acquiring new skills and knowledge without compromising its existing capabilities. Parametric methods suffer from catastrophic forgetting and loss of plasticity. Non-parametric methods, such as prompts, skills, and harnesses, leverage in-context learning but are limited by context length and lossy plain-text compression. To effectively combine their complementary strengths, we propose HERD (Harness-Evolving Recursive Distillation), a hybrid continual learning framework that learns efficiently in the harness and internalizes experience into cumulative capabilities via distillation. On the harness side, HERD continually refines its skills to capture reusable patterns, while retaining high-residual examples that the skills fail to capture in a bounded memory. On the parameter side, HERD performs drift-constrained distillation with dual anchors, substantially reducing forgetting over long recursive runs. On the 25-stage task stream of TRACE+17, under fixed memory and compute budgets, HERD matches the strongest parametric baseline in new-task acquisition while exhibiting the least forgetting and the strongest retention of general capabilities.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Forgetting, Leverage (statistics), Computer science, Machine learning, Context (archaeology), Artificial intelligence
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.31401, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange