EECS 195 / EECS 221

ML4SCI: Machine Learning for Science

University of California, Irvine · Fall 2026

Course description

What is machine learning for science, and how does it change the way we model, simulate, and discover the natural world? This course provides a broad introduction to the field, covering physics-informed learning, neural operators, generative and probabilistic models, and scientific foundation models, with particular emphasis on the representations, inductive biases, and evaluation criteria specific to physical systems and PDEs.

Instructor

Aparna Chandramowlishwaran Professor

Course contact: course email / Canvas announcement channel TBD

Logistics

Time: Tuesday & Thursday, 8:00–9:20 AM
Location: ICS 213
Office hours: By appointment (email to schedule)
Course numbers: EECS 195 (undergraduate) and EECS 221 (graduate)

Coursework

Students will be graded for participation in lectures and for their final project.

Notebook-guided classes: The course will be taught using notebook-guided lectures. Students are expected to attend and actively participate in class. Students should complete and submit the notebooks through Canvas.
Final project: Students will conduct an original investigation, replication, extension, or comparative study in machine learning for science.

Grading policy, deliverables, and submission instructions will be available to enrolled students through Canvas.

Schedule

The schedule is subject to change. Readings, class notebooks, and other materials will be added throughout the quarter.

Class	Date	Topic	Materials
Part I — What makes ML for Science different?
1	Thu, Sep 24	Introduction	Slides + Notebook
2	Tue, Sep 29	ML Foundations for Scientific Data	Slides + Notebook
3	Thu, Oct 1	PDE and Numerical Fundamentals	Slides + Notebook
4	Tue, Oct 6	Data Representations for SciML	Slides + Notebook
5	Thu, Oct 8	Symmetry, Invariance, and Equivariance	Slides + Notebook
Part II — Learning spatial and temporal dynamics
6	Tue, Oct 13	CNN and U-Net Architectures for Spatial Fields	Slides + Notebook
7	Thu, Oct 15	Autoregressive Rollout Models	Slides + Notebook
8	Tue, Oct 20	Temporal Modeling: Recurrent Networks and Neural ODEs	Slides + Notebook
9	Thu, Oct 22	Temporal Modeling: Attention and Transformers	Slides + Notebook
10	Tue, Oct 27	Evaluating SciML	Slides + Notebook
Part III — Changing the computational formulation
11	Thu, Oct 29	Operator Learning	Slides + Notebook
12	Tue, Nov 3	GNNs and Mesh-Based Learning	Slides + Notebook
Part IV — Incorporating scientific knowledge
13	Thu, Nov 5	Physics-Informed ML	Slides + Notebook
14	Tue, Nov 10	Physics-Constrained and Hybrid Learning	Slides + Notebook
Part V — From predictions to distributions
15	Thu, Nov 12	Generative and Probabilistic Models for Science	Slides + Notebook
16	Tue, Nov 17	Uncertainty Quantification	Slides + Notebook
Part VI — Scaling scientific intelligence
17	Thu, Nov 19	Scaling Laws for SciML	Slides
18	Tue, Nov 24	Mixture-of-Experts and Scientific Foundation Models	Slides + Notebook
19	Thu, Nov 26	No Class — Thanksgiving
19	Tue, Dec 1	Mechanistic Interpretability for Science	Slides + Notebook
20	Thu, Dec 3	Course Synthesis
Final	Tue, Dec 8 8:00–10:00 AM	Final Project Presentations	Project Presentations

Frequently asked questions

Who is this course for?

The course is intended for senior undergraduate and graduate students interested in applying modern machine learning to scientific and engineering problems. The undergraduate and graduate offerings share a common syllabus, with additional expectations for graduate enrollment.

What background is expected?

Students should be comfortable with core machine learning concepts and programming. Familiarity with linear algebra, probability, and differential equations is helpful. The course will introduce the scientific computing and PDE concepts needed to motivate the methods we study.

Is this a course about AI agents or LLMs for science?

Not quite. The course focuses on the machine learning foundations of scientific modeling: representations, inductive biases, dynamics, operators, physical constraints, uncertainty, scaling, and interpretability. Scientific foundation models appear later as one part of the broader landscape.

Will the course be useful if I work outside physics?

Yes. Although PDEs and physical fields provide important running examples, the principles extend to many scientific domains, including climate, materials, biology, chemistry, and other settings with multiscale data, governing constraints, and dynamics.

What will the final project look like?

Students will formulate and investigate a substantive question in machine learning for science. Projects may involve reproducing and critically evaluating a recent paper, extending an existing method, comparing modeling approaches, or applying modern methods to a scientific dataset.

Can we work in groups?

Students are encouraged to work in groups of 2-3.