Bridging Text
"Bridging" in scientific research refers to the integration of disparate data modalities or concepts to improve model performance or understanding. Current research focuses on bridging text and other modalities (images, sound, code, tabular data) using various techniques, including multimodal VAEs, diffusion models, and large language models (LLMs) adapted for specific tasks like automatic speech recognition or computer-aided design. This work aims to improve efficiency, accuracy, and explainability in diverse applications ranging from medical image analysis and robotics to legal reasoning and social science. The ultimate goal is to create more robust, versatile, and human-centered AI systems.
Papers
Bridging Model-Based Optimization and Generative Modeling via Conservative Fine-Tuning of Diffusion Models
Masatoshi Uehara, Yulai Zhao, Ehsan Hajiramezanali, Gabriele Scalia, Gökcen Eraslan, Avantika Lal, Sergey Levine, Tommaso Biancalani
TAMBRIDGE: Bridging Frame-Centered Tracking and 3D Gaussian Splatting for Enhanced SLAM
Peifeng Jiang, Hong Liu, Xia Li, Ti Wang, Fabian Zhang, Joachim M. Buhmann