SJTU Deng Lab is affiliated with Qing Yuan Research Institute, Shanghai Jiao Tong University, working on pioneering novel methods and theories that enhance the efficiency, efficacy, reliability, and trustworthiness of machine learning approaches, especially for generative AI. Our members conduct comprehensive research across the entire machine learning cycle, encompassing data, methods, models, and evaluation.

Current research interests focus on:

• Model efficiency

  • Inference acceleration for large language and diffusion models
  • Novel methods and architectures for generative modeling

• Data efficiency

  • Data selection, active learning, and data difficulty quantification
  • Deep spectral methods

• Modality efficiency

  • Unified modeling methods, learning principles, and architectures for texts, images, videos, and beyond
  • Applications of efficient world models in Embodied AI