两篇新的arXiv论文探讨了新兴的模型合并领域，该领域可以在无需访问原始训练数据的情况下组合独立训练的神经网络。第一篇论文介绍了用于单任务和多任务设置的C$^2$M$^3$和MERGE$^3$等算法，为组合学习能力提供了理论基础。第二篇论文研究了影响合并成功的因素，确定梯度对齐指标是兼容性的关键指标，并提出了面向合并的微调策略。 AI
影响 开发了组合和重用AI模型能力的基础技术，有可能降低训练成本并提高模型的多功能性。
排序理由 两篇在arXiv上发表的学术论文介绍了模型合并的新算法和分析。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.01580v1 Announce Type: new Abstract: Modern deep learning usually treats models as separate artifacts: trained independently, specialized for particular purposes, and replaced when improved versions appear. This thesis studies model merging as an alternative paradigm: …
arXiv:2601.22285v5 Announce Type: replace Abstract: Model merging combines knowledge from separately fine-tuned models, yet the factors driving its success remain poorly understood. While recent work treats mergeability as an intrinsic property of the models, we show with an arch…