Machine Learning (ML) based methods have shown great potentials to provide cost-effective solutions for building performance improvement and building-grid integration. This seminar introduces the state-of-the-arts of typical ML based data-driven smart building applications, such as ML based model predictive control, fault detection and diagnosis, and occupant centric control. The seminar will discuss the gaps and challenges of applying ML based methods in real buildings, such as data quality and method scalability. The seminar will also provide an overview of International Energy Agency’s ANNEX 81 (Data-driven Smart Buildings) as well as other on-going projects that the speaker is engaged in.