Final Exam
The Fall 2025 Final Exam served as an assessment of the advanced generative architectures and language models presented in the second half of the semester. This exam was non-cumulative and strictly covered material taught after the midterm exam.
Status: Concluded
The Fall 2025 Final Exam was successfully held on February 5, 2026.
Logistics (Archived)
- Date: Wednesday, ۱۶ بهمن ۱۴۰۴ (February 5, 2026)
- Time: 09:00
- Location: Department of Mathematical Sciences
Scope & Material
The exam covered Topic 9 through the end of the course.
- Topic 9: Energy-Based Models
- Topic 10: Score-Based Models
- Topic 11: Flow Matching
- Topic 12: Diffusion Models
- Topic 13: Evaluation of Generative Models
- Topic 14: Parameter-Efficient Fine-Tuning (PEFT)
- Topic 15: Multi-Modal Models
- Topic 16: LLM Emergent Abilities
- Topic 17: RAG and RLHF
Scope Note
Since this exam is not cumulative, we recommend focusing your study time entirely on the advanced architectures (e.g. Diffusion, Flow Matching) and the theoretical underpinnings of Large Language Models. While foundational knowledge from the first half of the course (e.g., probability, basic neural networks) is assumed, you will not be tested on specific pre-midterm architectures like VAEs, GANs, or Normalizing Flows.
Preparation Advice
The lecture slides cover the complete scope of the exam, including all necessary mathematical derivations. However, given the concise nature of the slides, we highly recommend consulting the Supplementary Material linked on the Course Material page. These resources provide the full context and narrative needed to gain a complete and deep understanding of the concepts presented in class.
Previous Exams
You can review the final exam from the previous offering (Fall 2024 or 1403-1) to practice your timing and familiarity with the question style.
Important Note on Curriculum Changes
Please note that this year's curriculum (Fall 2025) included new topics that were not present in the Fall 2024 syllabus. Therefore, the previous exam below does not cover these newer concepts and should be used only as a partial reference.