arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.07145cs.LG
收藏

失败是注定的,但可以被淡化:在大规模视觉和语言模型中表征和缓解 unwanted 行为

Failures Are Fated, But Can Be Faded: Characterizing and Mitigating Unwanted Behaviors in Large-Scale Vision and Language Models

School of Computing and Augmented Intelligence, Arizona State University, Tempe, United States of America(计算与增强智能学院,亚利桑那州立大学)

Som Sagar, Aditya Taparia, Ransalu Senanayake

AI总结:

本文提出利用深度强化学习表征和缓解大规模视觉和语言模型中的失败模式,通过有限的人类反馈重构失败景观,验证了方法在计算机视觉、自然语言处理和视觉-语言任务中的有效性。

摘要:

在大型深度神经网络中,尽管在许多任务上表现惊人,但也观察到与准确性、社会偏见和与人类价值观的一致性相关的失败。因此,在部署这些模型之前,必须表征这种失败景观,以便工程师调试和立法机构审计模型。然而,无法穷尽所有可能导致模型失败的因素组合。本文介绍了一种事后方法,利用深度强化学习探索并构建预训练判别和生成模型的失败模式景观。借助有限的人类反馈,我们展示了如何通过远离发现的失败模式来重构失败景观。我们实证地展示了所提出方法在常见计算机视觉、自然语言处理和视觉-语言任务中的有效性。

英文摘要:

In large deep neural networks that seem to perform surprisingly well on many tasks, we also observe a few failures related to accuracy, social biases, and alignment with human values, among others. Therefore, before deploying these models, it is crucial to characterize this failure landscape for engineers to debug and legislative bodies to audit models. Nevertheless, it is infeasible to exhaustively test for all possible combinations of factors that could lead to a model's failure. In this paper, we introduce a post-hoc method that utilizes \emph{deep reinforcement learning} to explore and construct the landscape of failure modes in pre-trained discriminative and generative models. With the aid of limited human feedback, we then demonstrate how to restructure the failure landscape to be more desirable by moving away from the discovered failure modes. We empirically show the effectiveness of the proposed method across common Computer Vision, Natural Language Processing, and Vision-Language tasks.

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。

↑