Kjartan Gudmundsson

Papers

Total Citations

H-Index

About

Kjartan Gudmundsson is a researcher whose work lies at the intersection of machine learning, Bayesian inference, and semi-supervised learning. His most notable contribution, "Deep Bayesian Self-Training" (2020), has garnered 36 citations, reflecting its significance in advancing methods that combine deep neural networks with probabilistic reasoning to improve learning from limited labeled data. This work introduces a principled framework for self-training that leverages Bayesian uncertainty, enabling models to make more reliable predictions and effectively leverage unlabeled examples—a critical challenge in many real-world applications where annotated data is scarce. Gudmundsson’s research has practical implications for fields ranging from computer vision to natural language processing, where data annotation is expensive or time-consuming. His approach stands out for its rigorous theoretical grounding and demonstrated empirical performance, offering a robust alternative to traditional self-training techniques. By bridging Bayesian statistics and deep learning, Gudmundsson has contributed to a growing body of work that seeks to make AI systems more data-efficient and trustworthy. His efforts, supported by collaborations with annotators and reviewers, underscore a commitment to both methodological innovation and reproducible, impactful science.