π AI and Materials Databases for Self-driving Labsο
Unleash the power of data science in the realm of self-driving laboratories. This remote, asynchronous course empowers you to apply data science concepts to materials discovery tasks. Youβll create Bayesian optimization scripts, explore advanced optimization topics, and adapt templates to create an advanced optimization setup for a materials discovery task. Topics will include multi-objective, constrained, high-dimensional, multi-fidelity, batch, asynchronous, and domain-aware Bayesian optimization. Additionally, youβll learn to share your findings by uploading datasets to a data repository, creating benchmark models, and hosting models on data science platforms.
Animation of 1D Bayesian Optimization: A Gaussian Process surrogate model can be used with an acquisition function to seamlessly transition from exploration to optimization in noisy settings. Source: https://ax.dev/docs/bayesopt.html
π Recommended Prerequisitesο
The recommended prerequisite for this course is Introduction to AI for Discovery using Self-driving Labs
π― Learning Outcomesο
Describe and categorize a materials discovery task using data science language and concepts
Customize a Bayesian optimization script to systematically identify the optimal chocolate chip cookie recipe, demonstrating practical application of optimization techniques
Evaluate and select an advanced optimization setup that is best suited for a specific materials discovery task, showcasing critical analysis and decision-making skills
Develop and execute a program to upload a dataset to a public database, construct a benchmark model, and deploy it online, illustrating proficiency in data sharing and model hosting
π οΈ Competencies/Skillsο
Data science literacy
Bayesian optimization
Advanced Bayesian optimization
Workflow orchestration
Benchmarking
𧩠Modulesο
Each module is intended to take approximately 3-4 hours, assuming that the recommended prerequisites have been met.
Module Name |
Topics |
Learning Outcomes |
---|---|---|
Intro to Bayesian optimization |
|
|
Mechanics |
|
|
Multi-objective optimization |
|
|
Constrained optimization |
|
|
High-dimensional optimization |
|
|
Batch/asynchronous optimization |
|
|
Featurization |
|
|
Multi-fidelity/multi-task |
|
|
Benchmark datasets/models |
|
|
βοΈ Course Assessments and Grading Schemaο
Each student is required to complete various quizzes and GitHub Classroom assignments. The course is structured into an orientation module followed by six subsequent modules. The course is graded on a pass/fail basis with 70% as the threshold for passing. Here is the breakdown of the points for each part of the course:
- π§ Orientation Module: Worth 15 points.
- π Modules 1-6: Each includes:
- π§ A guided notebook tutorial (ungraded)
- π A knowledge check (graded, 5 points)
- π οΈ A GitHub Classroom assignment (graded, 10 points*)
*The final module's GitHub Classroom assignment is worth 30 points.
Note that partial points are available on certain assignments.
π€ Course developer(s)ο
Sterling Baird, PhD Materials Science and Engineering (Acceleration Consortium)