Multiple Instance Learning

释义 Definition

多示例学习（MIL）：一种机器学习设定/方法。训练数据不是单个样本，而是由多个样本组成的“包（bag）”。通常只给每个包一个标签：例如正包表示包里至少有一个样本为正；负包表示包里所有样本都为负。常用于弱监督场景，如医学影像、目标检测、文本与音频片段识别等。

发音 Pronunciation (IPA)

/ˈmʌltɪpəl ˈɪnstəns ˈlɜːrnɪŋ/

例句 Examples

Multiple instance learning is useful when only bag-level labels are available.
当只有“包级别”的标签可用时，多示例学习很有用。

We trained a multiple instance learning model to identify malignant regions in whole-slide images, even though the annotations were provided only for each slide as a whole.
我们训练了一个多示例学习模型来在全切片图像中识别恶性区域，尽管标注只提供到了每张切片整体的层面。

词源 Etymology

“Multiple instance learning”由 multiple（多个）+ instance（实例/样本）+ learning（学习）构成。该术语在机器学习研究中用于描述一种“以样本集合（包）为单位监督”的学习框架，强调训练信号来自集合层面而非单个样本层面，因此常被归入弱监督学习相关方法。

相关词 Related Words

• Weak Supervision
• Bag
• Instance
• Bagging
• Positive Label
• Negative Label
• Latent Variable
• Attention
• Pooling
• Semi-supervised Learning

文献与著作中的用例 Literary / Notable Works

• Dietterich, T. G., Lathrop, R. H., & Lozano-Pérez, T. (1997). Solving the multiple instance problem with axis-parallel rectangles.
• Maron, O., & Lozano-Pérez, T. (1998). A framework for multiple-instance learning.
• Andrews, S., Tsochantaridis, I., & Hofmann, T. (2002). Support vector machines for multiple-instance learning.
• Zhou, Z.-H., & Zhang, M.-L. (2002). Neural networks for multiple-instance learning.
• Carbonneau, M.-A., Cheplygina, V., Granger, E., & Gagnon, G. (2018). Multiple instance learning: A survey of problem characteristics and applications.