Projects

We’ve flipped the script on multimodal fusion, instead of forcing visual features into discrete text space, we map text embeddings into continuous visual space, eliminating costly alignment pre-training while achieving competitive performance.

A fusion network for integrating Graph Neural Network and LLMs to generate text-related edge features in social network via knowledge injection and PEFT.

AI for coding transcripts and AI for simulating real-world negotiation scenarios.

An vision transformer application for ancient mortars classification.