arXiv cs.AI· Gyutae Oh, Hyunmin Kong, Jitae Shin·· 10 小时前AI 评分22
Lamarck:面向无回放域增量学习的继承式适配器专家
Inherited Adapter Experts for Replay-Free Domain-Incremental Learning
AI 导读
Lamarck 用免训练的 Mahalanobis 路由器为每张测试图像选择专家,并通过在每个 ViT 深层模块后校正图像 token 的小型适配器增强专家,在 CDDB-Hard 上最终准确率提升 4.4 点。
正文 · 原文
Abstract:Replay-free domain-incremental learning trains a model on the same classes in successive domains without keeping earlier images. Per-domain learners freeze a pre-trained ViT and train an additional expert for each new domain: the small set of parameters and the classification head learned on that domain. Test images carry no domain label, so a router selects one expert for each image. Recent work trains this router. However, a training-free Mahalanobis router costs the analyzed learners under 2.7 points of final accuracy relative to always selecting each image's own-domain expert. The experts, not this router, limit accuracy. Our learner, Lamarck, therefore uses this router and strengthens the experts. Its adapters, small trainable layers that correct every image token after each deep ViT block, add 4.4 points of final accuracy on the deepfake benchmark CDDB-Hard. In addition, each expert is initialized from its predecessor at no extra cost: the domains share their classes, so an image misrouted to a later expert loses less. Against twelve frozen-backbone baselines under one protocol, Lamarck is first on CDDB-Hard by 5.2 points, second on DomainNet and OfficeHome, and highest in mean accuracy.
| Comments: | 21 pages (8 pages main text, plus references and supplementary material), 8 figures, 23 tables |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.24145 [cs.CV] |
| (or arXiv:2611.24145v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2611.24145 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Gyutae Oh [view email]
[v1]
Wed, 18 Nov 2026 09:53:15 UTC (607 KB)
来源:arXiv cs.AI · arxiv.org