Solving challenging visual pattern recognition problems with deep convolutional neural networks: applications in biomedicine and robotics

The talk presents a few recent applications of deep learning techniques to extremely challenging visual pattern recognition problems, whose satisfactory solution would have been considered impossible until a few years ago. In particular, we focus on three tasks: two in the field of biomedicine (neural membrane segmentation in electron microscopy image stacks; mitosis detection in breast cancer histology images), and one in the field of robotics (detection and following of forest trails).
We show how approaches based on convolutional nets significantly outperform alternative techniques, and often reach human-level performance.

Alessandro Giusti is a Researcher at the Dalle Molle Institute for Artificial Intelligence in Lugano, Switzerland. He was awarded a PhD at Politecnico di Milano in 2009. His research focuses on robotic perception, biomedical imaging, visualization, and applications of machine learning.