Seminar: Explainable AI

Some information on this page is provisional and subject to change.

AI is currently taking the world in storm. Every day there are new news articles that comment on how it will profoundly change our lives or discusses the dangers humanity is in due to AI. This seminar will tackle one specific topic Area in this discussion: Explainable AI. 
Explainability methods aim to increase the transparency of AI models by showing why or how they arrive at their decision.

Students will be asked to address at least two AI Explainability methods, analyze them, implement them for practical evaluation, assess the methods’ strengths and weaknesses, and propose improvements. 

Given the extra effort involved, this seminar is compensated by 6 ECTS.

Module (Bachelor): Seminar: Explainable AI (BSc) (03SM22BIS023)

Module (Master): Seminar: Explainable AI (MSc) (03SM22MIS023)

• Abraham Bernstein
• Oana Inel
• Athina Kyriakou
• Kathrin Wardatzky

The following information is provisional and subject to change.

English

Prior Knowledge

It is expected that you are familiar with basic AI techniques such as the ones taught in the course "Introduction to Artificial Intelligence" on the BSc level.  

We also expect a familiarity with the typical programming environments in today’s AI landscape (mostly the python toolchains including cython) as well as typical software engineering tools such as git.

Software libraries

We will provider some base software libraries developed at IfI based on Cornac for your development.

Support

We will assign a buddy (i.e., fellow student) to you, with whom we heavily suggest you have code walkthroughs to ensure code quality and receive feedback.

We advise your to have a weekly meeting withy our buddy.

The following information is provisional and subject to change.

The followin is the basis for grading:

• Two versions of a seminar paper with associated software code and evaluation (first readable version, final version)
• Implementation of two AI explainability methods (possibly based on existing code)
  • exact information will be provided in the opening session.
• Participation in the 360-degree feedback: qualified feedback session to your peers for >= 3  other seminar papers ("Peer-Review")
• Continuous and active participation in the block seminar at the end of the semester, including a presentation of your seminar paper.

Additional information

• Assessment criteria (intermediary report, final report, peer review, presentation, implementation): will be provided in opening session
• Passing criteria: will be provided in opening session

The following information is provisional and subject to change.

• This is an in-person seminar. So in-person participation is mandatory.
• Materials and recording will be made available (most probably via Teams; possibly via OLAT or gitlab). The Seminar list limited to 20 students: 14 from the M.Sc. and 6 from the BSc program
• Note that, sadly, we will mosty probably not be able to provide any compute ressources 
• As of fall 2022, the allocation of students to seminars at the Department of Informatics will be done through the new module booking tool. More information on the process can be found on our website: Seminar allocation in Informatics
• DEADLINE for Registration: 04.02.2024 till 24:00

Will be provided in the opening session