SRISHTI YADAV

ELLIS PhD Fellow - University of Copenhagen and University of Amsterdam, member Pioneer Centre of AI and ILLC.

I am an ELLIS PhD Fellow advised by Dr. Serge Belongie (University of Copenhagen) and Dr. Ekaterina Shutova (University of Amsterdam). I am broadly interested in AI and Society. More specifically, I am currently working with multimodal models on problems of:

Alignment
Cross cultural awareness

For longer bio, click here.

Note: If you are planning to apply for ELLIS PhD program, feel free to send me questions about our lab and/or the program.

Latest Updates:

2024/11/30: Our paper 'Cross-modal Information Flow in Multimodal Large Language Models' is now up on arxiv. We examine the information flow between different modalities -- language and vision -- in Multimodal Large Language Models. One of the surprising finding was that for critical information from image to question, the two different critical information flows are observed in different window sizes where both occur in lower layers and sequentially follow each other.
2024/10/31: Our survey paper 'Survey of Cultural Awareness in Language Models: Text and Beyond' has been published. The papers talks extensively about cultural alignment in language, vision-language, HCI, social science, ethics of using large models for cultural alignment.
2024/09/26: Our paper 'Revealing Fine-Grained Values and Opinions in Large Language Models' has been accepted at EMNLP 2024.
2024/05/16: Organized Fine Grained Visual Categorization Workshop at CVPR 2024 as Competitions Chair

Rant: I used to have a fancy website but hated dependency on ever changing web frameworks. Back to basics. HTML is cool. Even Berkshire Hathaway uses it.