对比多种无标签表示质量度量，发现内在维数最可靠。
A Comparative Study of Label-free Representation Quality Metrics in Deep Learning
- 按构造方式分三类，分析同类度量间的关联性
- 在260个视觉模型上验证，内在维数预测能力最强
- 度量可靠性受网络结构和训练目标影响
我们对深度神经网络中无标签表示质量度量进行了对比研究，以评估其在多种配置下的可靠性。将现有无标签度量分为三类，基于其构建方式并建立同类度量间的理论联系。通过受控的合成实验刻画谱度量的敏感性。最终在六大数据集（涵盖通用物体分类、细粒度分类、场景识别、地理空间任务）上的260个视觉模型上，评估所有度量与下游任务准确率的关系，并按模型架构和训练目标进行分层分析。结果表明，在所考察度量中，内在维数（ID）是最可靠的预测指标。然而，包括ID在内的所有度量的可靠性均受模型架构和训练目标的影响。研究为理解无标签度量的实际含义、适用条件及使用方式提供了更清晰的依据。
原文摘要 · Abstract (English)
We present a comparative study of label-free metrics for assessing the quality of representations in deep neural networks to understand their reliability under a wide variety of configurations. We group existing label-free metrics into three families based on their construction and analytically establish connections between metrics within the same family. We then characterise the sensitivity of spectral metrics through controlled synthetic experiments. Finally, all label-free metrics are evaluated against downstream task accuracy across a diverse set of 260 vision models on six datasets spanning generic object classification, fine-grained object classification, scene recognition and geospatial task, stratifying results by architecture class and training objective. We find that intrinsic dimensionality (ID) is the most reliable predictor among the metrics considered. However, the reliability of all metrics, including ID, is moderated by architecture class and training objective. Our results provide a clearer understanding of what label-free representation quality metrics measure, when they are reliable, and how to interpret them in practice.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。