Artificial Intelligence
对比学习
Contrastive LearningA self-supervised learning paradigm that encourages a model to learn meaningful representations by pulling together similar instances and pushing apart dissimilar ones within a high-dimensional space. / 一种自监督学习范式，通过在高维空间中拉近相似实例并推远不相似实例，鼓励模型学习有意义的表示。
- Knowledge type
- 数学理论
- First proposed or discovered
- Proposer
- Turkey Farm member
- Creator
- 农场主
- Last updated
- 2026-11-13 21:06:35
Sub-knowledge points
Masked Image Modeling / 数学理论Hyperspherical boundary-guided learningUncertainty-rectified learningZero-prior label evolution
Higher-order knowledge points
Self-supervised learning / 数学理论Knowledge Distillation / 工艺技术Representation LearningDeep Metric Learning
Main parameters and performance indicators
Temperature: Controls the sharpness of the softmax distributionProjection Head: A neural network layer used to map data to a latent space
Open problems
Scalability to large-scale datasetsHandling complex data modalitiesMitigating the impact of noisy labels
Related literature
ranked by community starsURDN: Hyperspherical boundary-guided uncertainty-rectified contrastive learning for Railway Point Machine diagnosis under zero-prior label evolution and imperfect sensingOpen source
Related products and links
No item recorded yet
Creation and modification trail
1 recordsRecord title: Contrastive Learning.
Node update feedback
review required