Contrastive learning is a machine learning technique that creates an embedding space where similar data points are grouped together and dissimilar ones are separated. This method can be applied in both supervised and unsupervised settings, offering advantages over traditional cross-entropy loss functions, particularly in safety-critical applications. Research indicates that supervised contrastive learning can lead to more trustworthy and transparent neural networks by improving feature attribution explanations. AI
排序理由 The cluster contains two arXiv papers discussing contrastive learning techniques and their properties.
- Chopra et al.
- Contrastive Loss
- Cross-Entropy
- FaceNet
- Gutmann
- Hyvarinen
- Lifted Structured Loss
- Lilian Weng
- Noise Contrastive Estimation
- N-pair Loss
- Schroff et al.
- Sohn
- Supervised Contrastive Learning
- Triplet Loss
- Contrastive Representation Learning
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
