Prof. Kwanghoon Sohn’s team proposed Video+X LLM, which enables pre-trained video LLMs to process new modality data, and saliency-aware model merging, which integrates task-specific models into a unified foundation model. Two works have been accepted to ICML 2026 (H5-Index: 272).

These works were led by Dr. Jungin Park and Prof. Kwanghoon Sohn