[ML] An evaluation of data analysis techniques in digital health applications
With the surge in data availability in healthcare, the potential of data-driven digital health applications rises. In this research you will review different categories of digital health applications, and investigate the suitability of different data analysis techniques to digital health applications. Finally, you will benchmark these different techniques on a real-world medical dataset in the context of proving the effectiveness of a digital health application. You will contribute to showing which techniques are most effective for specific types of digital health applications. Daily supervisor: Jim Achterberg (LUMC), Marco Spruit[ML] Balanced and balancing distance measures for mixed variable types
Many AI, ML and data science methods depend on the notion of a distance, which often acts as a dissimilarity measure between observations in the data set. In real-world data sets, variables have various types, e.g. continuous, ordinal, nominal/categorical and binary, contained within one data set. In such cases, dissimilarity is almost always measured using Gower's distance. It min-max-scales numeric variables, and assigns distances to non-numeric variables as 1 if the values are unequal, and 0 if they are. Dimensions are just added directly, like in the Manhattan distance measure. The implication is that distances are dominated by categorical dimensions, as the distance (if non-zero) corresponds to the largest possible distance in the numeric dimensions, which will typically have smaller values. Also, average distances per dimension are not equalized (not even if the dimensions themselves are normalized or standardized first), and are dominated by imbalanced columns. This project will develop a balanced version of Gower's distance that makes the contribution of every feature on average equal, and leaves the possibility to re-weigh the contribution of features. The resulting distance measure will be used for risk stratification of people with metabolic syndrome on a large scale data warehouse with health, demographic and socio-economic data, but is expected to find wide-spread use in distance-based machine learning tasks on heterogeneous data.
Daily supervisor: Marcel Haas (LUMC), Marco Spruit[NLP] From mobile app to furry social robot: Welzijn.AI
Welzijn.AI is a new digital solution for monitoring mental wellbeing in the elderly. Currently in the form of a mobile app, its ultimate aim is to be embedded within a furry cat toy through a speech-only interface.
In this thesis topic you will investigate how speech-supported apps can be transformed into embedded/ambient technology devices. You start from a Raspberry-PI approach to embed within a furry cat or other toy, on the intersection of interative computing, affective computing, NLP, LLM and Health Informatics.
[NLP] Dutch NLP with English BERT models
It has been shown that fine-tuning English BERT models on translated Dutch clinical text can achieve results comparable to using Dutch BERT models on the original Dutch text. Given that Dutch BERT models are often trained on smaller datasets, translating into English to take advantage of larger and more robust English BERT models presents an exciting opportunity. Interestingly, this translation approach is very unexplored in current NLP research, which typically focuses more on building new models for each language. A shift to translation could save tremendous time and resources. In this project, you will get the chance to research how well translation works on a broader range of Dutch NLP tasks, with the opportunity to expand it to other languages as well. Your work could play a key role in shaping the future of BERT research for minority languages, and it's an opportunity to make a very meaningful impact in the field of non-English NLP!
Based on Extracting Patient Lifestyle Characteristics from Dutch Clinical Text with BERT Models. Daily supervisor: Hielke Muizelaar, Marco Spruit