AiNed XS grant awarded to Joren Brunekreef
![image](https://aiforoncology.nl/media/news_photos/fomoshift.png.700x350_q85_upscale.png)
The Dutch Research Council has announced that they have awarded one of their National Growth Fund AiNed-XS grants to Joren Brunekreef, a postdoctoral researcher in the AI for Oncology group. Joren's project proposal, called "FOMO-Shift", aims to tackle the problem of distribution shift with the help of foundation models.
The accuracy and reliability of AI models can decrease when such models are applied to datasets that are slightly different from those that were used for training the model. This is a well-known issue that is commonly referred to as "distribution shift". For example, an AI model that was trained to detect tumors in MRI scans originating from one hospital may not perform equally well on scans that were acquired in another hospital that uses different MRI scanning equipment. My FOMO-Shift proposal centers around the development of a method that can partially address this problem of distribution shift. To this end, I aim to leverage the internal data representations that are generated by so-called "foundation models": neural networks that have been pre-trained on large and diverse datasets, without the need for labels. The core idea of the proposed method is to find transformations that can undo the distribution shift in these internal data representations. If the approach is successful, it may enable foundation models to transfer their knowledge more reliably between different datasets. One of the many potential applications of the method would be to allow for safe deployment of medical AI foundation models across different institutions, without requiring additional labeled data to account for the potential difference in the data acquisition protocols.