arXiv：cs.LG· Josh McGiff, Salma Mekaoui, Robert Shanahan, Nikola S. Nikolov·· 4 小时前AI 评分38
AutoAdapt：自动域发现实现低成本扩展
AutoAdapt: Automatic Domain Discovery Enables Low-Cost Extensibility
AI 导读
AutoAdapt 是一个模块化框架，通过自动发现潜在领域并独立并行训练各领域的 LoRA 适配器，无需修改其他适配器即可引入新领域和数据。在 14 个领域基准和 GPT-4o 成对评判中，AutoAdapt 达到了与全领域 LoRA 适配器相当的性能，且无需全模型重训练。该方法通过构造方式避免了领域间干扰，实现无分类体系的模块化领域专精，且无聚合性能损失。
正文
Abstract:Instruction-tuned models are deployed into environments where domains are heterogeneous and evolve, yet adding new domains or data typically requires costly retraining. We present AutoAdapt, a modular framework that incorporates new domains and data via targeted single-adapter training without modifying other adapters. The framework automatically discovers latent domains, uses them to train per-domain Low-Rank Adaptation (LoRA) adapters independently in parallel and performs parameter-free routing. Across 14 domain-specific benchmarks and GPT-4o pairwise judgements, AutoAdapt achieves parity with a LoRA adapter trained on all domains without requiring full-model retraining. We also find evidence of specialisation effect convergence across independent discovery methods. Overall, training each adapter on its own domain prevents domain interference by construction, thus enabling modular, taxonomy-free domain specialisation without aggregate performance loss or full model retraining.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.10349 [cs.LG] |
| (or arXiv:2610.10349v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.10349 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Josh McGiff Mr [view email]
[v1]
Wed, 7 Oct 2026 16:26:57 UTC (96 KB)
来源：arXiv：cs.LG · arxiv.org