Germany
- 9 years of experience applying AI to computer vision - 6 years of hands-on experience working with medical data - Industry experience across the full lifecycle: data collection, model development, and deployment; tech lead on two projects with a global pharmaceutical company - Research background with publications at leading venues such as ICCV and MICCAI - Team leadership experience managing up to 4 direct reports, including hiring, onboarding, performance feedback, and offboarding
As a data scientist at AIgnostics I participate in applied research projects as well as practical engineering tasks in the field of digital pathology. The focus lies on AI models for histopathological images to detect and classify cancer.
Within the European CLARIFY project I was working on AI applied to medical images. When doctors analyze a medical image they look for certain patterns and characteristics to perform a correct diagnosis. At the same time pattern recognition and image analysis is exactly the strength of deep learning. My motivation was to bring these two fields closer together! Two major problems of medical AI applications today are: 1. The lack of data - The data annotation of medical images requires expert knowledge and is very expensive. 2. The reliability of deep learning methods - Often it is not well understood, how deep learning methods work and when you can rely on the predictions. My PhD offered me the possibility to search for scientific solution for these problems. The goal was to tackle them with probabilistic deep learning and crowdsourcing methods. With a Bayesian modeling we can account for the different expert levels of the annotators. Additionally probabilistic deep learning models offer a possibility to estimate the uncertainty of predictions so that we know when we can trust a prediction - and when not. The focus of the project lied on the detection of cancer in histopathological images. AI can help doctors to find and classify the images correctly - to be able to prescribe the correct treatment and in the best case, save lives!
As a software engineer at TomTom I had the possibility to gain a lot of theoretical and practical experiences in the field of deep learning. My main areas of work were related to: - the semantic segmentation of street scenes - the classification of traffic signs - the creation of datasets - the anonymization of images My role as a software engineer offered incredible possibilities about how to design and maintain software in a professional way. This included the scrum workflow and CI/CD (continuous integration/continuous delivery) with many different tools, such as Jenkins, Mlflow, Sonarqube, Kubernetes, github and Bitbucket.
Scientific project with 360° cameras, especially the stitching of frames. Programming language: C++
Noise analysis and simulation of a satellite IMU sensor. Language: Matlab/Simulink