feature representations
Feature representations are the numerical encodings of inputs that models use as data points for learning patterns. Good feature representations capture relevant information and structure across the data, enhancing the model's ability to make predictions.
- AnaCP: Toward Upper-Bound Continual Learning via Analytic Contrastive Projection
- How Classifier Features Transfer to Downstream: An Asymptotic Analysis in a Two-Layer Model
- MaxSup: Overcoming Representation Collapse in Label Smoothing
- MaxSup: Overcoming Representation Collapse in Label Smoothing
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers
- OOD Detection with Relative Angles
- Scaling Image Geo-Localization to Continent Level
- Single GPU Task Adaptation of Pathology Foundation Models for Whole Slide Image Analysis
- X-Mahalanobis: Transformer Feature Mixing for Reliable OOD Detection