Develop cutting edge technologies for large scale industrial landing of large pre-trained models, with related topics covering embodied intelligence, AI agents, and multimodal AI.

Established the Fundamental and Deep Learning (FDL) team, consisting of members from renowned universities, research institutes, and corporate research centers both domestic and international. Led the team in conducting research in robust deep learning, large language models (LLMs), and autonomous decision-making for dynamic open-world scenarios, and deep learning-assisted scientific computing. A series of research achievements, such as the LLM4Teach framework for training RL agents more efficiently with guidance from LLMs, the when2ask algorithm for efficient interaction between pre-trained LLMs and specialized smaller agents, open-world few-shot learning algorithms that leverage collaboration among multiple foundation models, and the Q-Net deep learning network for few-shot image segmentation, have been showcased in top international conferences and publications in the field of AI, including NeurIPS, AAAI, IJCAI, IEEE TCVST, Neurocomputing, etc. These works have also gained attention and citations from world-class AI research institutions such as MIT CSAIL, CMU, Stanford University, Google Brain, and Deepmind. His team achieved two runner-up awards in world-class AI algorithm competitions. Bin has also been elected as a member of the Intelligent Computing Innovators in China for 2023 issued by MIT Technology Review China & DeepTech. Their codes are open-sourced at https://github.com/ZJLAB-AMMI. For papers, see https://scholar.google.com/citations?user=UWCOeegAAAAJ&hl or https://www.researchgate.net/profile/Bin-Liu-120.

Focusing on the company's e-commerce business, he conducted research and development of advanced AI algorithms (such as deep learning, Bayesian methods, and causal inference) for precise and personalized product recommendations and marketing.

Conducted academic research in the fields of Bayesian inference, machine learning, optimization methods, nonlinear state filtering, etc. Propose the theory of Bayesian dynamic multimodel ensembling (see review paper entiled "Robust sequential online prediction with dynamic ensemble of multiple models: A review" at https://www.sciencedirect.com/science/article/pii/S0925231223006768), derivative-free optimization methods based on sequential Monte Carlo (SMC) methods, robust particle filtering algorithms, and adaptive annealing importance sampling algorithms for model selection and posterior distribution sampling, among others.

Did research on machine learning, Bayesian methods with applications to problems in computational neuroscience.

Focusing on the decoding problem of HiSilicon's self-developed chip baseband communication module (a multiple-input multiple-output system), did research on advanced decoding algorithms. Developed a new decoding algorithm that combines matrix QR decomposition and Markov Chain Monte Carlo sampling and implemented it on the full-link simulation platform. It was verified to achieve optimal decoding accuracy. Due to this achievement, Bin won the Chip-Star Award from the Semiconductor Chip Algorithm Department.

Led the algorithm team to research machine learning, optimization, and experimental design algorithms for automated design of microstructures in metamaterials. Applied these algorithms to the design and production of practical metamaterial products.

Systematically researched Bayesian statistics, particularly sequential Monte Carlo methods and other Bayesian computational techniques, Bayesian nonparametric modeling, and model selection methods. Collaborated with statisticians and astronomers to develop advanced Bayesian computational algorithms for the detection of exoplanets in extra-solar space that resemble Earth.