Regular courses for research postgraduate students admitted under HKU IDS mainly fall into the following three inter-related categories: 

1. Foundation: what are the main problems that Data Science and Machine Intelligence try to address. 
2. Computation: how do we develop effective and efficient methods to solve these problems. 
3. Application: why and how are these methods can be applied or customized to real-world applications.  

In short, the above course categories aim to introduce to our HKU IDS students a wide spectrum of problems and applicable methods in the world of data science and machine learning.

1. Foundation

• Fundamental structures, patterns, or models that are important for modern data science and machine intelligence: linear/nonlinear low-dim models, sequential or dynamical models, geometric and graphical models, and basic statistical or geometric properties of these models;
• Problems associated with how to estimate such models from data, the associated statistical problems and computational problems; and how to estimate them in different settings: supervised, unsupervised;
• How such models can be used for different purposes: classification, generation, prediction, and inference etc.

Below is the current list of courses with brief introduction: 

• Theoretical Foundation of Deep Learning; Data Mining (regression/PCA for linear models, clustering and classification for mixture models, and introduction to more complex models such as manifolds or graphical models and the associated estimation and inference problems).  
• High-Dimensional Data Analysis (mainly mixture low-dim linear models) 
• Foundation of Sequential Decision Making (sequential, dynamical models) 
• Statistical Inference and Machine Learning for Network Data (graphical or other structured models)

2. Computation

Courses in this category show when and how we can develop both effective (correct) and efficient (scalable) methods, including both algorithms and systems, for solving the above problems at scale.  

Below is the current list of courses with brief introduction: 

• Optimization for Statistical Learning (sample complexity for the estimation problems to be well-defined, tractable, and even globally optimal, computational complexity in terms of scalable oracles, etc.) 
• Scalable Optimization Methods in Data Science (first-order convex or nonconvex, distributed and parallel, stochastic etc.) 
• Theoretical Foundation Foundations for Deep Learning (deep networks as unrolled optimization, deep learning as fine-tuning via back-propagation, etc.)

3. Application

Courses in this category teach the students how to apply fundamental models and computational methods to real-world domain-specific data and problems, including but not limited to visual data, natural languages, dynamical data, biological and medical data, financial data, as well as multi-modal data.  

Below is the current list of courses with brief introduction: 

• Advanced Deep Learning for Computer Vision (a graduate counterpart for the undergraduate version on deep learning) 
• Embodied AI for Robotics: Perception, Learning, and Action  
• Advanced Natural Language Processing  

What Courses to Expect in the Future 

In the future, courses for other topics that are important for Data Science and Machine Intelligence will be offered, subject to the instructors’ availability. Such courses include but are not limited to, large-scale computational systems, data security and privacy, explainable AI, applications in medicine, applications in financial data, applications in social science, and many others. 

HKU IDS seminars 

HKU IDS students are expected to attend a minimum of 10 seminars organized by the Institute per year.