Scholarlink

It is well known that machine learning models are susceptible to out-of-distribution (OOD) data, which occurs when the distribution of data at test time differs significantly from that at training time. This inherent shift in distribution signifies that we cannot guarantee that the resulting models can perform well when deployed in real-world scenarios. This issue has spurred numerous studies on OOD learning, aiming to mitigate the negative impacts raised by such distribution shifts.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Trustworthy Machine Learning with Out-of-Distribution Data

  • 章节

  • 发表日期：

    2026年

查看全文
  • Bo Han,
  • Tongliang Liu

摘要

It is well known that machine learning models are susceptible to out-of-distribution (OOD) data, which occurs when the distribution of data at test time differs significantly from that at training time. This inherent shift in distribution signifies that we cannot guarantee that the resulting models can perform well when deployed in real-world scenarios. This issue has spurred numerous studies on OOD learning, aiming to mitigate the negative impacts raised by such distribution shifts.