I'm an AI researcher focusing on Deep Learning, Explainable AI, and Computer Vision. I have a PhD in Machine Learning from Luleå University of Technology, Sweden. My thesis focuses on using pretrained CNNs to evaluate the similarity of images and how to use such similarity metrics to train other machine learning models. I currently work as a postdoctoral researcher at Umeå University in Sweden where I explore Explainable AI methods for images and beyond.
I'm a strong believer in open science and believe that in the field of Computer Science, the good practice is to release your research code for anyone to easily pick up and furhter develop your methods. As such I try to release repositories containing the code of my work along with any papers I publish, you can find my contributions (in reverse order of appearance) below. Each repository also links to the free versions of the research papers they have been used in (you can find the published versions by searching for the name of the work).
- Analysis of Deep Perceptual Loss Networks: This repository contains code that I worked on together with a few other researchers at Luleå University of Technology. It contains code for creating perceptual loss networks from many different pretrained CNNs in the Torchvision package and code for automatically evaluating them on benchmarks based on four prior works in the field. This repository was used in the works Deep Perceptual Loss and Similarity, Deep Perceptual Similarity is Adaptable to Ambiguous Contexts, and A Systematic Performance Analysis of Deep Perceptual Loss Networks: Breaking Transfer Learning Conventions.
- Analysis of Deep Perceptual Similarity: This repository contains code that I and another researcher developed at Luleå University of Technology. It contains code for evaluating several different methods for calculating deep perceptual similarity on the BAPPS dataset. This repository was used in the works Deep Perceptual Loss and Similarity and Identifying and Mitigating Flaws of Deep Perceptual Similarity Metrics.
- Improving Autoencoders with Deep Perceptual Loss: This repository contains code that I developed as Luleå University of Technology to analyze how image autoencoder training could be improved using deep perceptual loss. This repository was used in the works Deep Perceptual Loss and Similarity, Deep Perceptual Loss for Improved Downstream Prediction, Pretraining Image Encoders without Reconstruction via Feature Prediction Loss, and Improving Image Autoencoder Embeddings with Perceptual Loss.
- Know Your Intent: This repository contains code developed by me and a few other researchers spread across multiple institutions. It contains code for training and evaluating a host of different machine learning methods on intent classification using subwords semantic hashing. This repository was used in the work Subword Semantic Hashing for Intent Classificationon Small Datasets.
Additionally, while I did not participate in the development of the following repositories, I did contribute to the papers based on them:
- VidHarm: This repository contains code for training and evaluating audio-video models for automated film age rating. This repository was used in the work VidHarm: A Clip Based Dataset for Harmful Content Detection.
- Magnification Prior for Self-supervised Learning: This repository contains code for training image models using Self-supervised Learning with image magnification as "distortion". This repository was used in the work Magnification Prior: A Self-Supervised Method for Learning Representations on Breast Cancer Histopathological Images.