About

The MSRA Shanghai AI/ML Group focuses on machine learning and its applications. More specifically, we are interested to the following directions.

AI and brain

We are interested in the interdisciplinary study of AI and the brain, with the goal of using AI to enhance our understanding of the brain and applying these insights to improve both AI and brain health.

  • Brain-inspired AI: This includes spiking neural networks, brain-inspired neural network architectures, and brain-inspired algorithms, among other areas.
  • Brain-computer interfaces: Our focus is on understanding EEG signals and developing non-invasive EEG-based BCI applications.
  • Embodied AI: Our aim is to develop new techniques that enable agents to behave flexibly in complex environments with open-ended goals.

AI and Healthcare

We are interested in applying cutting-edge machine learning techniques to real-world healthcare problems, aiming to use AI to enhance the productivity of doctors and medical researchers.

  • AI-aided speech therapy: We propose deep learning-based algorithms to detect and estimate hypernasality in patients with cleft lip and palate.
  • AI-based drug discovery: We develop new drug combination algorithms for more effective cancer treatments and design innovative diffusion algorithms for flexible molecule generation.
  • Medical image analysis: We aim to improve the performance of ultrasound localization microscopy in medical ultrasound videos and develop effective pretraining and finetuning techniques for medical image foundation models.

AI and Knowledge Discovery

We are focused on explainable machine learning to help people better understand AI models and to enhance human learning by unveiling the hidden knowledge within today’s advanced AI systems.

  • Explainable machine learning: We propose new techniques for understanding time series models, graph neural networks, and computer vision models.
  • AI-aided knowledge discovery and learning: We leverage cutting-edge AI models to facilitate human learning and to discover new knowledge from AI models.

LLM-based Agents

We are interested in leveraging LLM-based agents to solve real-world problems.

  • Task automation: We developed one of the first frameworks to utilize large language models (LLMs) for automating the solution of various tasks via connecting AI models.
  • Task planning: We explore the limits of LLMs in task planning and propose new algorithms to enhance their task planning capabilities.
  • Benchmarks and datasets: We released TaskBench, one of the largest datasets for task automation and planning problems.