Petrus Mikkola
Scholar

Petrus Mikkola

Google Scholar ID: 9MxBELsAAAAJ
University of Helsinki
Machine learningBayesian statisticsDeep learning
Homepage↗Google Scholar↗
Citations & Impact
All-time
Citations
209
 
H-index
5
 
i10-index
3
 
Publications
20
 
Co-authors
0
 
Contact
CVOpen ↗GitHubOpen ↗
Publications
5 items
Uncertainty-Normalized Margins for Direct Preference Optimization
2026
Cited
0
Maximally Robust Satisficing Bayesian Optimization
2026
Cited
0
Score-Based Density Estimation from Pairwise Comparisons
2025
Cited
0
Normalizing Flow Regression for Bayesian Inference with Offline Likelihood Evaluations
2025
Cited
0
Non-geodesically-convex optimization in the Wasserstein space
arXiv.org · 2024
Cited
2
Resume
Background
  • Machine learning researcher
  • Choice theory and Bayesian enthusiast
  • Research focuses on probabilistic deep learning, especially learning from human data
  • Combines random utility theory with modern probabilistic models (e.g., flows, diffusion models) to represent beliefs of oracles such as humans and LLMs
  • Key interests: flows, elicitation (knowledge/prior elicitation, preference learning), and Bayesian optimization