How I Study AI - Learn AI Papers & Lectures the Easy Way

Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models

Intermediate

Xiaomin Yu, Yi Xin et al.Feb 2arXiv

This paper finds a precise way to describe and fix the Modality Gap, which is when image and text features that mean the same thing still sit in different places in the AI’s memory space.

#Modality Gap#Multimodal Large Language Models#Contrastive Learning

Papers1

Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models