Conceptembedding analysis
Embedding Analysis
A method for assessing representation quality by analyzing the geometric distribution of samples in a model's embedding space. Common metrics include class compactness and class separation, and it is used to understand the feature structure learned by the model.
Timeline (last 90 days)
Oct 19
Embedding analysis showed that diseases with fewer real training images tended to gain greater class separation, but geometric changes and classification performance did not show a consistent correlation.
Unverified50%