English

If sufficient demand

The aim of the course is to explore modern generative modeling approaches, with a particular focus on diffusion models, transformer architectures, large-scale architectures based on self-supervised learning and reinforcement learning from human feedback (RLHF), and parametric and implicit models for visual reconstruction and synthesis.
A secondary aim is to further develop critical thinking and scientific creativity, enabling students to identify open problems and formulate new research directions beyond immediate incremental developments. The course emphasizes methodological understanding, conceptual clarity, and research judgment in rapidly evolving and computationally demanding areas of generative AI

The course covers the following topics:

● Diffusion models, transformer architectures, and static and temporal generative models

● Parametric models and implicit functions for reconstruction and synthesis, 3D visual modeling

● Generative and multimodal architectures based on self-supervised learning and reinforcement learning from human feedback (RLHF)

Modern developments in deep learning and AI are driven both by large-scale empirically developed models and by rapidly advancing fundamental methodologies. The course adopts a critical methodological perspective, examining how to balance modeling complexity, empirical
scale, theoretical structure, and experimental sufficiency under increasing computational, data, and organizational constraints.

The course is intended for students who are already familiar with standard deep learning and generative modeling techniques and focuses on frontier-level concepts, open problems, and emerging research directions.

Advanced research literature on diffusion models, transformer architectures, and spatial and temporal generative models

Parametric and implicit modeling approaches for visual reconstruction and synthesis

Current research frontiers, limitations, and evaluation practices in large-scale generative modeling

Critically analyze advanced generative AI literature, including large-scale modeling and optimization principles underpinning modern architectures

Synthesize insights across multiple research directions and assess methodological trade-offs

Formulate coherent and original research questions and modeling directions grounded in identified gaps and limitations of existing approaches

Independently and critically explore novel research directions in generative AI

Have ability to critically evaluate and reflect on research literature and identify open questions and promising research directions in generative AI.

Reason about methodological trade-offs, research ethics, and feasibility in the development of research ideas and exploratory research proposals.

Lectures

Seminars

Self-study literature review

Oral exam

Written report

An individual or group essay and oral presentation on a selected topic, where the student/group expands on feasible future research directions. The examination is conceptually similar to developing a research agenda or exploratory proposal, broader in scope than a single paper and aimed at identifying promising questions, approaches, models, or research programs in an important and insufficiently understood domain.
The examination is designed to assess the student/group’s ability to critically synthesize literature, formulate original research directions, and reason about feasibility, scope, and methodological trade-offs.

Failed, pass

The course is intended for advanced graduate students or other academics or professionals with a solid background in computer vision and machine learning. Prior research experience is expected; prior publications are beneficial but not required. The course is suitable for participants interested in deepening their research skills and engaging with frontier-level problems in generative AI.

 

Scientific articles and research papers covering both foundational work and the current research frontier in generative modeling, computer vision, and machine learning.

FMA355F

2026-01-15

/Jonas Johansson