Instructor: A. J. Medford
Email: [email protected]
Office Hours: By Appointment
Class Hours: W 10:10-11:00am
Office Hours Location: ES&T L1222
Class Room: MoSE 1201A
The Vertically-Integrated Projects (VIP) Program operates in a research and development context. Undergraduate students that join VIP teams earn academic credit for their participation in design/discovery efforts that assist faculty and graduate students with research and development issues in their areas of expertise.
The teams are:
-
Multidisciplinary - drawing students from all disciplines on campus;
-
Vertically-integrated - maintaining a mix of sophomores through PhD students each semester;
-
Long-term - each undergraduate student may participate in a project for up to three years and each graduate student may participate for the duration of their graduate career.
The continuity, technical depth, and disciplinary breadth of these teams are intended to: Provide the time and context necessary for students to learn and practice many different professional skills, make substantial technical contributions to the project, and experience many different roles on a large, multidisciplinary design/discovery team. Support long-term interaction between the graduate and undergraduate students on the team. The graduate students mentor the undergraduates as they work on the design/discovery projects embedded in the graduate students’ research. Enable the completion of large-scale design/discovery projects that are of significant benefit to faculty members’ research programs.
This course explores projects at the intersection of computational chemistry (quantum mechanics) and data science (big data) within the application domain of surface science and catalysis. The team merges expertise from computational and physical sciences, and students from computer science, electrical engineering, industrial & systems engineering, chemical engineering, chemistry, physics, and materials science. The VIP course consists of 3 classes of sub-team projects:
-
Training: All new students must complete a “training” project that involves the calculation of an adsorption energy using the quantum-mechanical technique of density functional theory (DFT), and training of a neural network model that can reproduce DFT results. The codes for DFT are well-established, but require the use of supercomputing resources and must be converged with respect to several numerical parameters. The training exercise will introduce students to high-performance computing, quantum-mechanical simulations, and machine learning packages used for atomic simulation data.
-
Data Generation: These projects will utilize DFT calculations performed on the PACE supercomputer to generate datasets of scientific relevance for problems in surface science and catalysis. This may involve the identification of new catalyst materials via high-throughput screening, or gaining an in-depth understanding of the catalytic behavior of a material based on modeling of various intermediate species and reaction pathways. Projects may also involve improving the codes used to generate the DFT data, or the infrastructure used to collect and organize the resulting data.
-
Machine Learning: These projects will focus on training machine-learning models to reproduce the output of DFT calculations. In particular, “neural network force-fields” provide a route to rapidly predict the energies and forces on atoms based on training data gathered from DFT or other atomistic simulation techniques. There are established packages like the Atomistic Machine-learning Package (AMP) and the Simple-NN package where these models are implemented. However, optimizing the architecture and parameters of the neural network remains challenging. These projects will work with specific datasets and students will try to improve the performance of the machine-learning model by varying hyper-parameters. Students may also explore new machine-learning algorithms and/or improve the codes and infrastructure used for neural network training.
The course will utilize the following resources for communication and submission of assignments:
-
Github: All course materials will be posted to Github.
-
Canvas: The course Canvas site will be used for submission of assignments and peer grading.
-
Slack: The group Slack channel is the preferred method of communication with instructors and graduate students. You are responsible for anything that is announced in the
vipSlack channel.
-
Calculate adsorption energies using the density functional theory (DFT)
-
Converge numerical calculations with respect to input parameters
-
Submit, manage, and analyze high-performance computing jobs
-
Utilize machine-learning packages to predict the output of atomistic simulations
-
Work with a team to solve real-world problems at the intersection of big data and quantum mechanical simulations
The grade will be assigned based on three categories:
-
Documentation: 33.3 %
-
Personal Accomplishments: 33.4 %
-
Teamwork and Interactions: 33.3 %
A grade of 0-5 will be assigned for each category based on the criteria outlined below. A total grade will be computed based on the weighted average of the 3 categories which will be converted to letter grades using the following:
-
A: > 4
-
B: > 3
-
C: < 3
-
D: < 2
You will receive a grade at the midterm and after the end of the course. The midterm grade is only advisory, and will not factor into the final calculation.
The documentation grade will be assessed based on the VIP notebook and additional documentation in the Github repository. All external documentation should be referenced in the VIP notebook. Please clearly mark any external sources by boxing them and highlighting, and/or placing them on a separate page. If you prefer to use an electronic notebook or digital tool for documentation it must be approved beforehand. There is a notebook grading rubric available on the course Canvas site that will be used to grade the notebooks. The notebook will account for a total of 3/5 points toward your documentation grade.
The documentation grade will also include contributions from two additional sources: the in-class updates and a literature review assignment. The slides from all in-class updates should be uploaded at the midterm and final. The literature review will be submitted to Canvas. The in-class updates and literature review will be worth 1 point each, and account for a significant portion of your final grade, so you should be diligent about ensuring they are submitted on time.
Personal accomplishments will be measured by self-defined goals and a combination of instructor and peer evaluation. Within the first week of the semester students returning to projects will define midterm and semester goals for their project. Each goal should have a deliverable that can be unambiguously evaluated as complete or incomplete (computer code, report, figure, etc.), and each student should submit a deliverable. Deliverables submitted by team members need not be unique, but ideally self-defined goals will be individualized, such that each team member has different deliverables that support a larger team-based goal.
-
5: Goals are completely achieved and additional progress has been made (A+)
-
4: Goals are completely achieved (A)
-
3: Goals are partially achieved (B)
-
2: Some progress has been made, but goals are not achieved (C)
-
1: No substantial progress is made towards goals (D)
The accomplishments will be graded by your peers, and confirmed by the instructor. Importantly, the midterm assessment is only advisory, and students have an opportunity to revise goals within 1 week of receiving the midterm assessment (revisions must also be approved). The final grade will depend only on the revised goals. This means that the grade is controlled by two factors: i) ability to plan research and set realistic targets and ii) ability to achieve goals and deliver on a plan.
For new projects (including the training project for new students) the instructor or sub-team advisor will provide the initial goals. The training project goals are based on results from prior semesters and are known to be reasonable; therefore, goals cannot be removed or revised to be less challenging. However, new project goals may be overambitious and students are encouraged to revise them at the midterm once they have a grasp of the long-term goals and scope of the project.
The accomplishments will be initially graded by multiple peers and confirmed by the instructor. The deliverables will be submitted via Canvas as a .zip file, and a copy of the project goals document should be included with the submission to aid the reviewers. If a template is provided this should be strictly followed to aid in analysis of results. Any external deliverables (e.g. websites, Github, etc.) should be clearly referenced. This can also contain any comments, instructions, or context that may be important for a grader. When grading other students’ assignments you should use the above 1-5 scale and criteria as a guideline. Only materials included in or referenced in the submission should be used to assess the criteria (e.g. if a group presents something in the VIP meeting but does not include it in the submission then it does not count).
Teamwork and interaction will be graded based on peer evaluations
conducted through the VIP website.
conducted through the VIP website and your participation in peer grading. Failure to complete peer grading or peer evals will result in 1 (out of 5) points deducted from the Teamwork and Interaction grade. If you do not complete either then you will lose 2 (out of 5) points.
Note that you only need to complete peer evals for your sub-team members, but peer grading will be across different subteams and will be explicitly assigned.
The first weeks will involve an overview of the team, available projects, and discussion of goals. Subsequent weeks will include training lectures by graduate students, and/or 10 minute update presentations by VIP sub-teams. The goal of these presentations is to present progress and challenges so that the rest of the class can make suggestions. Updates should be informal and consist of 5 or fewer slides. The remaining time in the class will be used as a “workshop” where teams can interact with their sub-team advisor and/or each other as needed. If nobody from your group is present when you are scheduled to present then everyone in the group will lose 1/2 point (out of 5 total) from your “Documentation” grade. Please let an instructor know if nobody can make it, and work with other groups to find a time to make up the presentation.
Week 1: Introduction to VIP and projects (Medford)
Week 2: Overview of literature searches (Medford)
Week 3: Intro to Python (Lei)
Week 4: Intro to ASE and adsorption energies (Comer)
Week 5: PACE and Bash scripting (Lei)
Week 6: Density functional theory calculations (Comer)
Week 7: Workshop
Week 8: Midterm Updates: Training Groups
Week 9: Atomistic machine learning: theory (Lei)
Week 10: Atomistic machine learning: application (Comer)
Week 11: Optional workshop (no official classes due to COVID)
Week 12: Update presentation: Machine Learning Groups (remote)
Week 13: Final Updates: Traning Groups (remote)
Week 14: Final Updates: Machine Learning Groups (remote)
The schedule and syllabus are subject to change. Given that this is a research course, changes are to be expected; however, we will do our best to notify you of any changes and implement them as fairly as possible.