Marius Aasan
Scholar

Marius Aasan

Google Scholar ID: i9mjbNcAAAAJ
University of Oslo
Machine LearningImagingProbabilistic Machine Learning
Homepage↗Google Scholar↗
Citations & Impact
All-time
Citations
13
 
H-index
2
 
i10-index
0
 
Publications
5
 
Co-authors
0
 
Contact
Emailmariuaas@ifi.uio.noGitHubOpen ↗
Publications
4 items
Suppressing Non-Semantic Noise in Masked Image Modeling Representations
2026
Cited
0
Differentiable Hierarchical Visual Tokenization
2025
Cited
0
Why Prototypes Collapse: Diagnosing and Preventing Partial Collapse in Prototypical Self-Supervised Learning
2025
Cited
0
SPoT: Subpixel Placement of Tokens in Vision Transformers
2025
Cited
0
Resume (English only)
Academic Achievements
  • - Publications:
  • - "Pixel-Level Predictions with Embedded Lookup Tables," published in Proceedings of the Symposium of the Norwegian AI Society 2025
  • - "A Spitting Image: Modular Superpixel Tokenization in Vision Transformers," published in Lecture Notes in Computer Science (LNCS)
Research Experience
  • - Master's Thesis: Investigated the role of Invertible Neural Networks (INNs) in inverse problems for imaging and proposed new methods to bridge the gap between densely connected and convolutional feed-forward architectures and INN/NF (Normalizing Flow) structures
  • - Research Focus: Bayesian data analysis, computational statistics, and image processing
Education
  • - Degree: Master of Science
  • - University: University of Oslo
  • - Program: Data Science
  • - Time: Joined the master's program in 2019
  • - Advisor: Odd Kolbjørnsen (Master's thesis advisor)
  • - Undergraduate: Mathematics and Informatics program (MAMI), coursework included mathematics, statistics, and computer science
  • - Bachelor's Thesis: Focused on fractals and imaging, supervised by Tom Lindstrøm
Background
  • - Research Interests: Bayesian paradigms and probabilistic machine learning techniques, particularly generative modeling, normalizing flows, invertible neural networks, and encoder/decoder models
  • - Areas of Interest: Attention mechanisms in vision transformers, graph neural networks, representation learning, and methods for incorporating contextual priors in deep learning
  • - Mathematical Interests: Representation theory, linear analysis, manifolds, and some differential geometry
  • - Applied Mathematics and Signal Processing Interests: Frame theory, information theory, generalized Fourier and wavelet analysis
Miscellany
  • - Hobbies: Music composition, audio production, procedurally generated art, and recreational reading on various topics in physics and philosophy
Co-authors
0 total
Co-authors: 0 (list not available)