Sonia Joseph

Into the Rabbit Hull: From Task-Relevant Concepts in DINO to Minkowski Geometry

Sparse CLIP: Co-Optimizing Interpretability and Performance in Contrastive Learning

At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization

From Noise to Narrative: Tracing the Origins of Hallucinations in Transformers

Too Late to Recall: Explaining the Two-Hop Problem in Multimodal Knowledge Retrieval