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 2505.23933cs.LGcs.AI
收藏

BIRD:基于表示结构蒸馏的行为诱导

BIRD: Behavior Induction via Representation-structure Distillation

Department of Computer Science(计算机科学系);UC Santa Barbara(圣巴巴拉大学);Department of Psychological & Brain Sciences(心理学与脑科学系)

Galen Pogoncheff, Michael Beyeler

AI总结:

提出BIRD框架,通过匹配师生模型的内部表示结构转移对齐行为,在图像分类分布外鲁棒性上最高提升16%准确率,并发现教师表示的三个属性可解释85%的迁移成功方差,为安全AI部署提供指导。

摘要:

人类对齐的深度学习模型展现出与人类价值观一致的行为,如鲁棒性、公平性和诚实性。将这些行为属性转移到基于不同任务或数据分布训练的模型上仍然具有挑战性:对齐行为在微调过程中很容易被遗忘,且收集能保留该行为的特定任务数据成本可能极其高昂。我们提出了BIRD(Behavior Induction via Representation-structure Distillation),这是一个通过将学生模型的内部表示结构与教师模型相匹配来转移对齐行为的灵活框架。在应用于图像分类的分布外鲁棒性问题时,BIRD优于微调、迁移学习和持续学习方法,鲁棒准确率比次强基线最高提升16%。即使教师模型在更简单的数据集上训练且规模比学生模型小25倍,它依然有效。在一项包含400多对师生模型的大规模研究中,我们表明教师模型表示的三个可解释且可计算的属性(即任务相关性、行为相关性和互补知识)能够解释迁移成功中高达85%的方差。这些见解为教师模型的选择和设计提供了实用指导。BIRD将小型、良好对齐的模型转化为可扩展的对齐种子,消除了在现实环境中部署安全AI系统的一个关键瓶颈。

英文摘要:

Human-aligned deep learning models exhibit behaviors consistent with human values, such as robustness, fairness, and honesty. Transferring these behavioral properties to models trained on different tasks or data distributions remains challenging: aligned behavior is easily forgotten during fine-tuning, and collecting task-specific data that preserves this behavior can be prohibitively costly. We introduce BIRD (Behavior Induction via Representation-structure Distillation), a flexible framework for transferring aligned behavior by matching the internal representation structure of a student model to that of a teacher. Applied to out-of-distribution robustness in image classification, BIRD outperforms fine-tuning, transfer learning, and continual learning methods, improving robust accuracy by up to 16% over the next strongest baseline. It remains effective even when the teacher is trained on a much simpler dataset and is $25 \times$ smaller than the student. In a large-scale study of over 400 teacher-student pairs, we show that three interpretable and computable properties of the teacher's representations (i.e., task relevance, behavioral relevance, and complementary knowledge) explain up to 85% of the variance in transfer success. These insights offer practical guidance for teacher selection and design. BIRD turns small, well-aligned models into scalable alignment seeds, removing a key bottleneck in deploying safe AI systems in the wild.

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。

↑