Hands-on in Applied Math
The entire course is built in a way that requires action by the students. They are guided through the process of reading a paper and assessing it. This guidance is also what motivated students to participate, because it is a support they rarely get but highly value. The assessment of scientific work is implemented through a review process in which each student prepares presenting one paper to their fellow group members.
This definitely required their engagement. Also, the fact that they had to decide on three accepted and two rejected papers made it quite natural to have a vivid group discussion. For the coding part, the students very much enjoyed the competitive nature of the process, with the leader board. This made it a fun experience to learn this new coding challenge of attacks and defenses.
A Course Roadmap
Introduction: overview of the subject and relevant theory given by the lecturer.
Paper reading: a guided process on learning how to read scientific papers and how to assess them with the help of the lecturer. The same two papers are read by all the students. They review the papers in form of a written assignment and then, discuss them in small groups. The students receive feedback on their written reviews.
Quality criteria: as a big group activity, quality criteria for scientific work are brainstormed by the students together with the lecturer. These are discussed, prioritized and grouped.
Peer review: In groups of five students, we mimick a conference review process. Each group assesses the same five papers and ranks them. The three best papers get “accepted» to the mock conference, the other two do not. Then, we gather all the groups and discuss the different rankings of the groups. Here is an example of the review process result.
In addition, 5 students volunteered to give short talks (15 minutes pitches) on the individual papers.
This way, those students who wanted could get experience in giving a presentation and at the same time, each student got an overview on all the papers, not just the one they read. After each presentation, students had the chance to give feedback to their fellow classmates as well as ask questions and participate in a discussion on the presented work.
Coding: assistants provide a coding tutorial and help students get started. With their assistance, the students could build and employ software for image classification and adversarial attacks of image classifiers.
Coding challenge on attacks and defenses: with the help of a software library, each student could build an image classifier that they consider robust to attacks. In total, three rounds of coding tasks were given to the students. They consisted of attacking the models of the other classmates in certain ways, as well as improving on their own model through defense strategies. For each round, the results were posted on a leader board and the coding was graded.
Discussion of the leader board and results as a big group activity together, where students can also share their experience of the attacks-and-defenses part. This is followed by a lecture on state-of-the-art and future research directions in the area of adversarial attacks and defenses.
Final submission: Each student is asked to create a short video summarizing their experience and findings. Videos are uploaded on a platform and made accessible to all students.
Feedback for Students
The students received feedback on their written assignments. They also received (immediate) feedback on their quality criteria in class, as well as on their presentations (for those who gave one) by both the lecturer and the students. The teaching assistants also provided quick help and feedback on the coding part during tutorial sessions, but also through a Moodle forum.
Encouraging Student Engagement
The entire course is built in a way that requires action by the students. They are guided through the process of reading a paper and assessing it. This guidance is also what motivated students to participate, because it is a support they rarely get but highly value. The assessment of scientific work is implemented through a review process in which each student prepares presenting one paper to their fellow group members.
This definitely required their engagement. Also, the fact that they had to decide on three accepted and two rejected papers made it quite natural to have a vivid group discussion. For the coding part, the students very much enjoyed the competitive nature of the process, with the leader board. This made it a fun experience to learn this new coding challenge of attacks and defenses.
Effects on Student Learning
Students specifically valued that through this course they learned how to read and assess scientific papers, something they did not know how to do before taking the course. I experienced for the first time that students could learn a variety of new elements without having a final exam. Learning turned out to be such a natural result of this course. I knew what they had learned without having to give them an exam.
Accompany the Student Learning Process
For the paper review process, we had a “practice run” where everyone read the same two papers and provided written reviews. This way, the students already received feedback based on the first two reviews and could implement this for writing their third review.
Through the big group discussion on quality criteria, I could really ensure that all students were engaged and were actively thinking about objective quality criteria. The discussion helped to set a common understanding of these criteria.
The coding part was very closely monitored with the help of three teaching assistants. They met with the students regularly and made sure that any issues were immediately resolved.
Project team
Seminar für Angewandte Mathematik
Rämistrasse 101
8092
Zürich
Switzerland
