Teaching
Deep Statistics: AI and Earth Observations for Sustainable Development
Together with Professor Xiao-Li Meng, Department of Statistics, Harvard University, I am teaching our course Deep Statistics for the fourth time at Harvard, Chalmers, and Linköping.
Deep statistics refers to statistical endeavors that go deeper than developing methods and applying them to solve problems in data science.
It explores and develops statistical theory and insights to contribute to the building of foundations for data science. The course provides a deep dive into statistical foundations and insights for multi-source, multi-phase, and multi-resolution learning, interwoven with case studies on using AI and Earth Observations (EO) for sustainable developments (e.g., global poverty).
Foundational issues are framed as inevitable trade-offs for data science: between data quality and quantity, between statistical and computational efficiencies, and between robustness and relevance of learning methods and findings.
Practical questions examined include handling messy and private data, drawing causal conclusions from AI-EO data, and translating scientific insights into policies.
Recommended Prep
A student should preferably have at least one of the following background:
- Foundational and theoretical proficiency: at the level of STATS 210, 211. (Strong interest in statistical theory and foundational issues.)
- Data analytical and computational skills: basic data science skills and being able use image data.
Solving STAT 110 Introduction to Probability
I have a deep fascination in statistics. It is a versatile thinking tool and to a large extent the the soul of sciences.
When I was a postdoc at Harvard, I audited Joe Bliztstein’s Statistics 110: Probability. In his book, he provides a couple of hundred problems, and I have found it useful to solve a selection of them. I am planning to solve all of them. For a subset of the problems I have solved, I have recorded my solutions on YouTube. My long term project is to record all my solutions.
Chapter 6 solutions
Chapter 7 solutions
Past teaching
I taught extensively during my academic journey, across both substantive and methodological topics. For example, I developed a course in economic sociology together with collaborators, which we taught mainly at University of Gothenburg, Sweden, and we wrote a book on economic sociology for a Swedish audience. For methodologically oriented course, I have led teaching in causal inference, multilevel modeling, mixed methods, and visualization. Additionally, I have supervised more than 30 master these projects, both in computer science and social science.