Deep learning has lead to huge advances in longstanding problems of computer vision, natural language processing, signal processing and robotics. However, deep networks may render dangerous wrong predictions when they process corrupted or faulty images. Pablo Marquez Neila, Senior researcher at the Artificial Intelligence in Medical Imaging (AIMI) laboratory, will develop an „Image data validation for AI systems“ that can identify whether an input image can be safely processed by a deep network or not. Goal is to create a comprehensive methodology on data validation that should enable a large number of industries to build more robust AI systems.
Medicine has witnessed very interesting Deep Learning (DL) developments in recent years, but the black-box nature of the analytic capabilities of large Neural Networks has also raised many concerns by scientists and doctors. In this context, Visual Question and Answering (VQA) methods that query the content of an image by means of an explicit text question, offer an exciting new pathway to discern the inner workings of DL. Raphael Sznitman, head of ARTORG's AIMI lab, is proposing the elaboration of new VQA methods that can reason about the content of medical images in similar ways to domain experts. His project “Visual Question Answering and Visual Turing Tests for Medical Imaging“ also sets out to design methods that can validate and verify non-bias of VQA systems.
Artificial Intelligence in Medical Imaging
