https://dasya.itu.dk/supervisor/veronika-cheplygina/ Recent content in Veronika Cheplygina on DASYA Hugo -- gohugo.io en-us © DASYA, IT University of Copenhagen Fri, 22 Nov 2024 00:00:00 +0000 https://dasya.itu.dk/for-students/proposals/current/vs-similarity/ Fri, 22 Nov 2024 00:00:00 +0000 https://dasya.itu.dk/for-students/proposals/current/vs-similarity/ Machine learning methods are often evaluated on benchmark datasets, in computer vision, medical imaging, NLP and other fields. In such evaluation, researchers often describe the data as being: representative, for example based on the distribution of ages of the patients mirroring the world population, similar, for example because both dataset contain pictures of animals diverse, for example because data was collected from multiple countries Recent research shows, the concepts of representativity, similarity, and diversity, are not strictly defined, and different notions might be used by different researchers. https://dasya.itu.dk/for-students/proposals/archive/yucl-medical/ Wed, 26 Jun 2024 00:00:00 +0000 https://dasya.itu.dk/for-students/proposals/archive/yucl-medical/ Spectral learning priority is a useful tool in analyzing a model’s focus during training, it describes how a model may understand a given image from the spectrum perspective. For example, to distinguish cats and tortoises, learning to recognize their shapes would be enough, such embedding will result in higher learning priority at low frequencies representing shapes; while learning to distinguish turtles and tortoises requires more effort to extract fine-grained features, leading to the corresponding learning priority shifting to higher frequencies. https://dasya.itu.dk/for-students/proposals/archive/ajs-debias/ Tue, 07 May 2024 00:00:00 +0000 https://dasya.itu.dk/for-students/proposals/archive/ajs-debias/ It has been observed that deep learning models are able to identify patient characteristics such as age, sex, and self-reported race with high accuracy from medical images such as chest x-ray recordings, even when medical doctors cannot. This raises the potential for such models to learn to (falsely) diagnose patients of different demographics differently, even if they present with the same disease characteristics. This may happen because, for example, some groups may tend to be misdiagnosed or underdiagnosed in the datasets using which such models are trained. https://dasya.itu.dk/for-students/proposals/archive/vc-med-nlp/ Thu, 22 Apr 2021 00:00:00 +0000 https://dasya.itu.dk/for-students/proposals/archive/vc-med-nlp/ Machine learning is used extensively in different applications, including medical imaging and natural language processing. As different types of data are involved, it is reasonable to assume that different methods are needed for each application. However, there are also opportunities in translating a method successful in one application, to the other application where it is not widely used. The goal of this project is to examine trends in recent conference papers for both fields, and identify relevant similarities/differences between the approaches used. https://dasya.itu.dk/for-students/proposals/archive/crowdsourcing-similarity/ Thu, 22 Apr 2021 00:00:00 +0000 https://dasya.itu.dk/for-students/proposals/archive/crowdsourcing-similarity/ Machine learning algorithms for skin lesion classification typically learn from images which have been labeled as malignant (for example, melanoma) or not. Such tasks can still suffer from overfitting due to limited dataset size. In other computer vision tasks, crowdsourcing labels has been effective, but the average person typically does not have the background to classify skin lesions. However, assessing the similarity of images might still be possible, even without such background.