Galvin Khara

According to our database, Galvin Khara authored at least 3 papers between 2024 and 2025.

Collaborative distances:

• Dijkstra number of five.
• Erdős number of four.

2025

Transformers Don't Need LayerNorm at Inference Time: Scaling LayerNorm Removal to GPT-2 XL and the Implications for Mechanistic Interpretability.

Luca Baroni, Galvin Khara, Joachim Schaeffer, Marat Subkhankulov, Stefan Heimersheim

CoRR, July, 2025

Robust image representations with counterfactual contrastive learning.

Mélanie Roschewitz, Fabio De Sousa Ribeiro, Tian Xia, Galvin Khara, Ben Glocker

Medical Image Anal., 2025

2024

Counterfactual Contrastive Learning: Robust Representations via Causal Image Synthesis.

Mélanie Roschewitz, Fabio De Sousa Ribeiro, Tian Xia, Galvin Khara, Ben Glocker

Proceedings of the Data Engineering in Medical Imaging - Second MICCAI Workshop, 2024