1 paper · 1 filter
Isaac Roberts, Petra Bevandic, Alexander Schulz +1
The paper proposes a model‑agnostic method that uses automatically discovered concept activation vectors to explain why two images are considered similar, by perturbing embeddings…