- Artificial intelligence
- Data Science
- Software engineering
- Organising and planning events
- E-sports(LCN)
- Softball
- Hackathons
Welcome!
My name is Tanja Crijns. I'm 26 years old and currently living in Utrecht, The Netherlands. I did a BSc in Artificial Intelligence and I just finished my MSc in Data Science with distinction. On this website you will find more information about my interests and projects that I am currently working on.
Interests
Board of study association CognAC 2014-2015
Radboud University, FNWI ©klokgroep
Softball team, Dames 1, Radboud Rangers
Projects
- Conversational assistance track TREC For my masters thesis, I studied how BERT could be used in a multi-turn artificial conversational setting. It was great learning experience to join a conference track and to be working with a state-of-the-art NLP technique on a complex problem.
- Legal relevance classification Legal relevance classification of dutch corpora. Near perfect performance in both the supervised and unsupervised approach. Suspicion of data leakage even after thorough examination of variables. The source code and research paper can be found on Github.
- Kaggle competitions I have participated in the Kaggle competition 'The Nature Conservancy Fisheries Monitoring' together with a team. The task was to automatically categorise images that may or may not contain certain types of fish. There are eight categories. We ended the competition in the top 6%. A github repository of our project can be found here.
- Microbleed detection For a medical imaging course, I researched the use of deep learning in the detection of cerebral microbleeds in Traumatic Brain Injury patients. We used a two-step deep learning approach; segmentation and classification. I co-authored the paper that was written on this topic and it was accepted at the SPIE Medical conference in 2017 (session 10).
We also participated in the Kaggle competition 'Planet: Understanding the Amazon from Space'. The objective here was to classify which ground types (forest, cultivation etc.) are present in patches of a satellite image. We approached the problem with deep learning techniques and finished the challenge in the top 12%. A github repository of our project can be found here.