Erik Winge

• Stokke, Mira Sofie & Winge, Erik (2026). God bruk av kunstig intelligens i jusstudiene. Jussens Venner. ISSN 0022-6971. 61(1-2), s. 1–11. doi: 10.18261/jv.61.1-2.1. Fulltekst i vitenarkiv Vis sammendrag

  Bruken av kunstig intelligens øker, men dårlig bruk av kunstig intelligens kan føre til dårlig læring. Artikkelen drøfter hvordan jusstudenter kan bruke kunstig intelligens på en måte som fremmer læring. Artikkelen gir en innføring i hvordan store språkmodeller virker, og hvordan de kan brukes på en god måte som fremmer tenkning i jusstudiet. Artikkelen avslutter med tre konkrete råd til jusstudenter om bruk av store språkmodeller. Artikkelen er særlig rettet mot studenter, men kan nok også være nyttig for andre som er interesserte i god bruk av store språkmodeller.

Se alle arbeider i NVA

• Winge, Erik (2025). An Introduction to running Large Language Models on Fox. Universitetsbiblioteket i Oslo, Universitetet i Oslo. doi: https:/uio-library.github.io/LLM-course/. Fulltekst i vitenarkiv
• Winge, Erik (2021). An Introduction to Programming for Lawyers. Universitetsbiblioteket i Oslo, Universitetet i Oslo. doi: https:/uio-cell.github.io/programming-for-lawyers/. Fulltekst i vitenarkiv Vis sammendrag

  An Introduction to Programming for Lawyers contains the lecture notes for the course “JUS5080 – Programming for Lawyers” at the University of Oslo. The aim of this course is to give a basic understanding of algorithmic thinking and computational literacy.

• Winge, Erik & Hauge, Eirill Strand (2020). An Introduction to Programming in Python. Universitetsbiblioteket i Oslo, Universitetet i Oslo. doi: https:/uio-library.github.io/programming/. Fulltekst i vitenarkiv

Se alle arbeider i NVA

• Winge, Erik (2018). Word embedding models as graphs : conversion and evaluation. Universitetet i Oslo. Fulltekst i vitenarkiv Vis sammendrag

  In this thesis, we make and evaluate procedures for converting between different lexical semantic representations. We implement three different methods for converting from vector space models to graph models: the threshold method, the kNN method and the variable-$k$ method. We also implement a single procedure for converting from graph models to word embedding models. Further, we do comprehensive evaluation of our conversion procedures and their results. We perform extensive intrinsic evaluation using several gold standard data sets. We also do extrinsic evaluation of our converted graph models. We show that our graph models perform well at two real-world tasks: word sense induction/disambiguation and hypernym discovery. The use of word embeddings has become widespread, in large part because of the increasing use of deep learning methods. However, training high quality word embeddings is time-consuming and requires large corpora. Therefore, an increasing number of pre-trained word embedding models have been made available, for instance by the Nordic language processing laboratory. These pre-trained word embeddings are useful in many natural language processing tasks. However, there are also many graph-based algorithms. We show that we can utilize high-quality word embeddings in graph-based algorithms by converting word embedding models to graphs.

Se alle arbeider i NVA