Natural and Artificial Intelligence Seminar
ÕIS MTAT.03.292, 3 ECTS
2025/2026 Autumn Semester
- Topic: Abstraction in Vision and Language Models: Inferences from deep learning, brain, art and reasoning
- Target audience: Master's and Doctoral students in Computer Science and other computing-related disciplines. Some background in Machine Learning and Artificial Neural Networks is required to participate in the discussions of this course.
- Seminars: Thursdays (14:15-15:45)
- Location: Delta center-2006 in Tartu
- Seminar led by: Tarun Khajuria (tarun.khajuria@ut.ee) from NAIL
Content
Main questions include:
- What is the role of correct information abstraction for intelligence?
- How do deep learning models and the brain abstract information for categorisation, content generation and reasoning?
- How can we see deep learning models of vision and language through the lens of information abstraction? What are the implications for model hallucination, reasoning and generalisation in these models?
Tentative list of papers: https://docs.google.com/document/d/11od_e-MkwUSAU_x31BJod7LitKWhcnVvG8zbZ5xMZ1Y/edit?usp=sharing
Contact me (Tarun, tarun.khajuria@ut.ee) if you want to present a paper outside this list
Learning Environment
- Seminars take place in person
To Pass
- Select and present a paper. The presentation will be peer reviewed.
- Write reviews for 3 papers. 1 for the paper you presented, 2 for papers presented by others.
- Prepare a test with 4-5 questions on the paper you present.
- Obtain 60 % from the tests. This also sets an attendance requirement as tests will take place in person before the start of the seminar.