Han Liu / Selected Teaching
NORTHWESTERN UNIVERSITYUnderstand
deeply.Build what’s next.
The next generation of AI begins with how we learn to think. Three courses connect mathematical rigor, frontier models, and the craft of solving real problems.
Mathematical depth. Research perspective. Practical judgment.COMPUTER SCIENCE × DATA SCIENCE
The teaching perspective
From first principles
to independent thinking.
These courses approach intelligent systems from three complementary directions: the mathematics that makes learning possible, the architectures that turn it into capability, and the evidence needed to solve an open-ended problem.
01 / Reason from first principles
Graduate foundationsCOMP_SCI 416 · STAT 435
Mathematical Foundations of Machine Learning
What makes a learning method work?
Develop the mathematical language behind modern machine learning through a problem-solving framework. Connect model architecture with mathematical principles that make learning algorithms possible.
- Transformer architecture
- Geometric deep neural network
- Tensor algebra
- Vector calculus
- Reinforcement learning
θt+1 = θt − η∇L(θt)
Use local information to improve a decision.
p(θ | D) ∝ p(D | θ) p(θ)
Update beliefs in the presence of evidence.
E[f(X)] = ∫ f(x)p(x)dx
Reason about outcomes under uncertainty.
Inside the course
How the course is taught
In-class problem solving and take-home assignments build fluency with mathematical reasoning. Lecture notes support the course; no textbook is required.
Preparation & enrollment
Designed for PhD students, with instructor permission available as specified by Northwestern. Cross-listed between Computer Science and Statistics.
The intellectual habitTurn an intuitive idea into a precise mathematical question.
02 / Understand the frontier
Advanced topicsCOMP_SCI 496
Advanced Topics
on Deep Learning
How does architecture become capability?
Examine the architectures behind large pretrained models, with a focus on language, sequence and structures. Connect the ideas inside these systems to recent advances and practical applications.
- Pretrained models
- Neural architectures
- Language
- Vision
- Audio
- Deep learning tools
LanguageContext & sequences
VisionImages & representations
AudioSpeech & signals
Inside the course
Learning focus
Understand the design of pretrained models and develop a working perspective on emerging deep learning research. Selected topics evolve with the field.
Preparation & tools
Familiarity with convolutional networks, LSTMs, attention, and a framework such as PyTorch or TensorFlow. Python is the course programming language.
The intellectual habitRead a model architecture as a set of ideas and design choices.
03 / Make the knowledge work
Undergraduate capstoneSTAT 390
Data Science
Capstone Project
What can the data actually tell us?
Bring the full data science toolkit to real, open-ended problems. Move from acquiring and organizing data to modeling, visualization, and analysis—making decisions when the problem is less tidy than a textbook exercise.
- Data acquisition
- Wrangling & organization
- Visualization
- Modeling & analysis
- Open-ended projects
FROM QUESTION TO EVIDENCE390
01
Frame the problemDefine the question and its context.
02
Build the datasetAcquire, clean, and organize.
03
Model & investigateExplore patterns. Examine assumptions.
04
Make the evidence clearVisualize, interpret, and communicate.
A workflow for turning data into a defensible answer.
Inside the course
The project experience
Work with teammates to scope a problem and develop a solution for stakeholders with data needs. Integrate technical skills with collaborative problem solving.
Course context
Listed by Northwestern as “Data Science Project,” the capstone draws on the undergraduate data science sequence. Consult the catalog for current prerequisites and enrollment details.
The intellectual habitTake ownership of the question, the analysis, and the conclusion.
An education for what comes next
Learn to ask better questions.
Build answers that hold up.
01 /
Derive with precision.
Make assumptions explicit. Understand the reasoning that connects a method to its result.
02 /
Think in architectures.
Look inside a system. Connect its components, learning objectives, and capabilities.
03 /
Let evidence lead.
Bring ideas into contact with data. Interpret results with care and communicate them clearly.
The next research question starts here.
Selected courses from Han Liu’s teaching at Northwestern. For current offerings, enrollment, and term-specific requirements, use the official course listings.
Software
Principal Investigator
- Presidential Early Career Award in Science and Engineering (PECASE)
- Alfred P. Sloan Fellowship in Math
- IMS Tweedie New Researcher Award
- ASA Noether Young Scholar Award
- NSF CAREER Award in Statistics Science
- David P. Byar Young Investigator Award
- AISTATS Notable Paper Award
- Umesh Gavasakar Best Thesis Award
- ICML Best Overall Paper (HM)
Flickr gallery
