arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2412.09308cs.LG
收藏

面向持续测试时自适应的动态提示分配与调优

Dynamic Prompt Allocation and Tuning for Continual Test-Time Adaptation

Shandong University of Finance and Economics(山东财经大学);Shandong University(山东大学)

Chaoran Cui, Yongrui Zhen, Shuai Gong, Chunyun Zhang, Hui Liu, Yilong Yin

AI总结:

针对持续测试时自适应中共享模型易引发域间干扰、灾难性遗忘的问题,提出动态提示分配与调优方法PAINT,通过查询机制判断样本所属域并分配对应可学习域专属提示,结合互信息最大化与结构正则化调优,在三个基准数据集上验证了有效性。

摘要:

持续测试时自适应(CTTA)是近期兴起的技术,旨在将预训练的源模型适配到持续变化的目标分布中,以匹配真实世界环境的动态特性。为缓解CTTA中的灾难性遗忘风险,现有方法通常引入显式正则化项来约束模型参数的变化,但这类方法无法从根本上解决灾难性遗忘问题——因为它们依赖单个共享模型在所有目标域上进行自适应,不可避免地会引发严重的域间干扰。本文提出可学习的域专属提示,引导模型适配对应的目标域,从而在一定程度上解耦不同域的参数空间。在目标样本缺乏域身份标识的情况下,我们提出了全新的动态提示分配与调优(PAINT)方法:该方法通过查询机制动态判断当前样本来自已知域还是未探索域,对已知域直接选取对应的域专属提示,对未见域则分配新的提示;随后结合互信息最大化与结构正则化完成提示调优。在三个基准数据集上的大量实验证明了PAINT方法在CTTA任务中的有效性,我们已在https://github.com/Cadezzyr/PAINT发布代码。

英文摘要:

Continual test-time adaptation (CTTA) has recently emerged to adapt a pre-trained source model to continuously evolving target distributions, which accommodates the dynamic nature of real-world environments. To mitigate the risk of catastrophic forgetting in CTTA, existing methods typically incorporate explicit regularization terms to constrain the variation of model parameters. However, they cannot fundamentally resolve catastrophic forgetting because they rely on a single shared model to adapt across all target domains, which inevitably leads to severe inter-domain interference. In this paper, we introduce learnable domain-specific prompts that guide the model to adapt to corresponding target domains, thereby partially disentangling the parameter space of different domains. In the absence of domain identity for target samples, we propose a novel dynamic Prompt AllocatIon aNd Tuning (PAINT) method, which utilizes a query mechanism to dynamically determine whether the current samples come from a known domain or an unexplored one. For known domains, the corresponding domain-specific prompt is directly selected, while for previously unseen domains, a new prompt is allocated. Prompt tuning is subsequently performed using mutual information maximization along with structural regularization. Extensive experiments on three benchmark datasets demonstrate the effectiveness of our PAINT method for CTTA. We have released our code at https://github.com/Cadezzyr/PAINT.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑