Detecting liver cancer with transfer learning
Transfer-learning case study focused on model comparison, training behavior, and practical architecture trade-offs.
ResNet50ConvNeXtPyTorchMedical imagingTransfer learningModel comparisonDeep Learning
Applied ML · Bayesian inference · Medical technology
Bio
I’m Harald Karlsen, a 5th year master’s student in Applied Physics and Mathematics at UiT Arctic University of Norway, specializing in medical technology, scientific computing, and artificial intelligence. My work focuses on machine learning, Bayesian inference, and statistical modeling, with a particular interest in building interpretable methods for medical and health-related data. Through coursework and research projects, I’ve worked with topics such as deep learning, image analysis, spatial and Bayesian modeling, uncertainty modelling and signal processing. I’m especially interested in combining rigorous statistical thinking with practical machine learning tools to solve real-world problems.
Tromsø, Norway
Projects
Classifying news headlines using BERT models
A case study on fine-tuning BERT for text classification, with a focus on model performance, interpretability, and practical deployment considerations.
BERTNLPPyTorchText classificationModel interpretabilityDeep Learning
Comparing GradCAM to SOTA image segmentation methods
A case study comparing GradCAM to state-of-the-art image segmentation methods for interpretability in medical imaging, specifically for identifying ROIs in liver cancer scans.
GradCAMImage segmentationMedical imagingInterpretabilityPyTorchXAIDeep Learning
Comparing INLA and Metropolis-Hastings for Bayesian inference
A case study comparing INLA and Metropolis-Hastings for Bayesian inference to find mortality rate of beetles as a function of dose-level of a gas.
INLAMetropolis-HastingsBayesian inferenceStatisticsRMonte Carlo Markov ChainRegression
Spatial modeling of larynx cancer mortality with BYM2 and INLA
A Bayesian disease-mapping case study using R-INLA and the BYM2 spatial model to analyze larynx cancer mortality across 544 districts in Germany, estimate covariate effects, and compare raw versus fitted relative risks on a map.
R-INLABYM2Spatial statisticsBayesian inferenceDisease mappingPoisson regressionR
Current work
I am currently completing a master’s degree in Applied Physics and Mathematics at UiT, specializing in medical technology and scientific computing. My recent work includes deep learning, image analysis, Bayesian spatial modeling, and predictive modeling for clinical applications.