Participation (20%)

Graded in Phase 2:

5% - Attend all Phase 2 discussions, and say something. (If you need to miss a discussion, inform me ahead of time and you can be excused.)

5% - Questions that are relevant to the paper/topic being discussed.

10% - Answering questions or extending discussions with your own input.

5% bonus - especially insightful/interesting questions and answers.

You should take notes and write down questions while you are reading papers so you'll easily find something to say.

Assignments (20%)

Rubric to be given in assignments.

Paper Presentation (10%)

5% - Skill: Speak confidently and directly at the audience; minimize staring at notes or slides. Slides should have clear text and professional organization.

5% - Content: Show that you can present difficult concepts understandably. Highlight the most important results and contextualize them in the research landscape.

Project (30%)

10% - Project Presentation. Graded as above.

10% - Novelty. Your work should be a meaningful attempt to push boundaries, even if it does not successfully do so.

10% - Quality. The report should be readable and concise, with helpful diagrams and figures. It should be written with academic rigor. The code should run easily on a different system.

Oral Exam (20%)

You will be asked about your project and the course material. You are expected to fully understand your project even if you didn't work on a portion, sans technical details. I will divide grades into four tiers here, with possible slight variations of these grades:

0% - You did not demonstrate any knowledge of your own work and the course material. Your other grades may be subject to an investigation.

12% - You understood your work and the material on a surface level, but could not answer questions about them correctly. You did not demonstrate a solid grasp of machine learning and security, but you made a genuine attempt.

20% - Your understanding of your work and the course was fully satisfactory. You demonstrated independent, critical thinking based on a strong understanding of machine learning and security.

25% - You demonstrated excellent knowledge of the course material and answered questions with academic rigor. You were able to work out mathematical problems as well as theorize on novel variations.

Questions during the oral exam will be similar to these:

• What would happen if you used an ensemble of these three approaches to solve the problem? How would you construct the ensemble, and what would happen to the TPR?
• Let's talk about the paper you presented. One suggested weakness of the paper is that it didn't collect data correctly. How would you fix that?
• Let's look at this paper. In this work they use an uncommon metric to validate their work. Is it justified? How does this metric compare to more common ones?