按年被提到
20262026 · 8 次2026
强化学习与智能体›深度强化学习与世界模型·arXiv:2601.19897 · cs.LG

Self-Distillation Enables Continual Learning

Idan Shenfeld, Mehul Damani, Jonas Hübotter, Pulkit Agrawal · 2026
开始阅读 ↗PDFalphaXiv 讨论
被提及 · 2 人首次 2026-01-31 · 最近 2026-06-27
2不同的人
5活跃月份
7讨论串
5被引(OpenAlex)
arXiv 摘要(候选层不写导读)

Continual learning, enabling models to acquire new skills and knowledge without degrading existing capabilities, remains a fundamental challenge for foundation models. While on-policy reinforcement learning can reduce forgetting, it requires explicit reward functions that are often unavailable. Learning from expert demonstrations, the primary alternative, is dominated by supervised fine-tuning (SFT), which is inherently off-policy. We introduce Self-Distillation Fine-Tuning (SDFT), a simple method that enables on-policy learning directly from demonstrations. SDFT leverages in-context learning by using a demonstration-conditioned model as its own teacher, generating on-policy training signals that preserve prior capabilities while acquiring new skills. Across skill learning and knowledge acquisition tasks, SDFT consistently outperforms SFT, achieving higher new-task accuracy while substantially reducing catastrophic forgetting. In sequential learning experiments, SDFT enables a single model to accumulate multiple skills over time without performance regression, establishing on-policy distillation as a practical path to continual learning from demonstrations.

候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。

HN 书评

推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及
评论teleforce · 2026-03

“This is possible but not for training but fine-tuning the existing open source models. This can be mainstream, and then custom model fine-tuning becomes the new “software development”. Please check out this new fine-tuning method for LLM by MIT and ETH Zurich teams that used a single NVIDIA H200 GPU [1], [2], [3]. Full fine-tuning of the entire model’s…”

评论teleforce · 2026-04

“It seems that self-distillation is the way to go for LLM. It's has shown recent as very efficicient and effective very recently in January this year by the team by MIT and ETH in their Self-Distillatuon Fine-Tuning (SDFT) [1],[2]. This paper is also their closest competitor named on-policy self distillation in the comparison table. I hope they keep the…”

Apple: Embarrassingly Simple Self-Distillation Improves Code Generation原评论 ↗
评论teleforce · 2026-04

“>Why domain specific LLMs won’t exist: an intuition >We would have a healthcare model, economics model, mathematics model, coding model and so on. It's not the question whether there ever will be specialized model, rather it's the matter of when. This will democratize almost all work and profession, including programmers, architects, lawyers, engineers…”

Why domain specific LLMs won't exist: an intuition原评论 ↗
评论teleforce · 2026-05

“Fun facts, this paper is cited by Simple Self Distillation (SSD) paper Apple [1],[2]. I think is a bad naming scheme due to the very common SSD namesake and the fact that it belongs to on-policy self-distillation [3]. But again according to the authors their proposed solution is simple because "SSD uses only temperature-shifted samples from the base model…”

Self-Distillation Enables Continual Learning [PDF]原评论 ↗
评论teleforce · 2026-05

“Please check the recent self-distillation work by MIT-ETH, UCLA and Apple [1],[2],[3],[4]. I've got a strong feeling that self-distillation is the second best things happened to LLM after transformer breakthrough. [1]Self-Distillation Enables Continual Learning [pdf] (25 comments): https://news.ycombinator.com/item?id=48165265 [2] Self-Distillation En…”

Claude Opus 4.8原评论 ↗
评论teleforce · 2026-06

“>True, but the capabilities and knowledge of that model are also frozen in time, so the value of that model declines over time. Correction: The capabilities and knowledge of that model can be improved via self-distillation, so the value of that model increases over time. This is where I think self-distillation is the main way forward, and probably the s…”

The gap between open weights LLMs and closed source LLMs原评论 ↗
被提交为 story · 2 次