Course:
ME 697: Advanced Scientific Machine Learning
Prerequisites: ME 539 (Introduction to Scientific Machine Learning) or equivalent.
Instructor: Prof. Ilias Bilionis & Dr. Alexander Alberts
Description
Scientific Machine Learning (SciML) is an interdisciplinary field that combines machine learning, physics, and high-performance computing. While traditional ML requires vast amounts of data, SciML utilizes physical laws—like differential equations and symmetries—to compensate for data scarcity. This course covers advanced techniques such as differential and probabilistic programming, operator learning, and the integration of physical principles into ML frameworks.
What you will learn
- Uncertainty Quantification: Quantify and propagate uncertainties through scientific models.
- Advanced Calibration: Calibrate models using incomplete, noisy, and multi-modal experimental data.
- Dynamical Systems: Build predictive models of partially observed dynamical systems from data.
- Physics-Informed ML: Embed physical knowledge in modern machine learning methodologies.
- GPU Implementation: Implement SciML methodologies on graphical processing units (GPUs).
- Technical Reporting: Write comprehensive reports combining scientific computing with modern ML approaches.
Course Logistics & Resources
Course Communication
- Brightspace Discussion Forums: The primary means for course lessons, assessments, and Q&A.
- Office Hours:Conducted via Zoom (schedule posted in the course)
- Feedback:Homework and projects are graded within one week of the deadline.
Required Tools
Jupyter Notebooks:: Τhe core environment for all activities and assignments.
Google Colab: Recommended for running code; Colab Pro is suggested for projects requiring GPU acceleration.
GitHub Repository: Access all course activities via the official Git version control repository.
Grading & Assessment
Grading Policy
Your course grade is calculated based on the following components:
- Quizzes (5%): Most lessons include required, graded quizzes based on the lesson content. You will have two attempts for each quiz, and only your highest score will be recorded.
- Homework (55%): There will be seven (7) homework assignments, including both theoretical proofs and computational tasks using Jupyter notebooks. Submissions are made through Gradescope.
- Individual Project (40%):Each student will carry out an individual project that aims to emulate the process of writing a research paper with scientific machine-learning content.
Grading Scale
- A+: > 98% | A: 88–98% | A-: 85–88%
- B+: 80–85% | B: 73–80% | B-: 70–73%
- C+: 67–70% | C: 62–67% | C-: 60–62%
- D+: 57–60% | D: 52–57% | D-: 50–52%
- F: < 50%
Teaching Assistant:
- bwassgre@purdue.edu
- Office Hours (ET): Mon: 3:30 - 7:00 PM , Thurs: 11:00 AM - 1:00 PM , Fri: 4:00 - 6:00 PM
