arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.01909cs.LGcs.AIcs.CY
收藏

关于大型基础模型的灾难性继承

On Catastrophic Inheritance of Large Foundation Models

Carnegie Mellon University(卡内基梅隆大学);Microsoft Research(微软研究院);William & Mary(威廉玛丽学院)

Hao Chen, Bhiksha Raj, Xing Xie, Jindong Wang

AI总结:

本文提出大型基础模型存在灾难性继承问题,即从有偏预训练数据继承的缺陷会损害下游任务,并设计 UIM 框架从理解、解释和缓解三方面应对,以促进负责任的 AI 发展。

摘要:

大型基础模型(LFMs)正展现出令人难以置信的性能。然而,不仅在机器学习领域,而且在其他多个学科中,人们对其神秘且未经解释的潜力产生了巨大担忧。在这篇立场论文中,我们提出识别一个深植于大型基础模型中被忽视的问题:灾难性继承,描述从有偏的大规模预训练数据继承到大型基础模型在下游任务上的行为的弱点和局限性,包括损坏、长尾、噪声、分布外等样本。这种继承可能对下游应用造成灾难,例如偏见、缺乏泛化能力、性能退化、安全漏洞、隐私泄露和价值错位。我们讨论了这一问题背后的挑战,并提出 UIM,一个框架,用于从预训练和下游适应两方面理解大型基础模型的灾难性继承,解释灾难性继承对下游任务的影响,以及如何缓解它。UIM 旨在联合机器学习和社会科学社群,以实现更负责任和更有前景的人工智能开发与部署。

英文摘要:

Large foundation models (LFMs) are claiming incredible performances. Yet great concerns have been raised about their mythic and uninterpreted potentials not only in machine learning, but also in various other disciplines. In this position paper, we propose to identify a neglected issue deeply rooted in LFMs: Catastrophic Inheritance, describing the weaknesses and limitations inherited from biased large-scale pre-training data to behaviors of LFMs on the downstream tasks, including samples that are corrupted, long-tailed, noisy, out-of-distributed, to name a few. Such inheritance can potentially cause catastrophes to downstream applications, such as bias, lack of generalization, deteriorated performance, security vulnerability, privacy leakage, and value misalignment. We discuss the challenges behind this issue and propose UIM, a framework to Understand the catastrophic inheritance of LFMs from both pre-training and downstream adaptation, Interpret the implications of catastrophic inheritance on downstream tasks, and how to Mitigate it. UIM aims to unite both the machine learning and social sciences communities for more responsible and promising AI development and deployment.

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。

↑