Wat we doen
We would like to invite you for the first Deep Learning Nijmegen Meetup, an initiative of the Nijmegen Data Science Centre.
In the first meetup two inspiring speakers, Marcel van Gerven (Radboud University) and Taco Cohen (Qualcomm/UvA).
Marcel van Gerven is professor of Artificial Cognitive Systems at Radboud University and principal investigator in the Donders Institute for Brain, Cognition and Behaviour. His research focuses on brain-inspired computing, understanding the neural mechanisms underlying natural intelligence and the development of new ways to improve the interaction between humans and machines. He is head of the Artificial Intelligence department at Radboud University.
Taco Cohen is a machine learning researcher at Qualcomm Research Netherlands and a PhD student at the University of Amsterdam, supervised by prof. Max Welling. He co-founded Scyfer, a successful deep learning services company, acquired by Qualcomm in 2017. He holds a BSc in theoretical computer science from Utrecht University and a MSc in artificial intelligence from the University of Amsterdam, and has done internships at Google Deepmind and OpenAI. His research is focussed on understanding and improving deep representation learning, in particular learning of equivariant and disentangled representations, and data-efficient deep learning.
The talks, including questions will take about 1.5 hours. After the event, our sponsors will provide drinks and snacks for our networking event.
*Abstract Marcel van Gerven*
In this talk I will provide a general overview of work pursued in the Artificial Cognitive Systems group. I will touch upon some recent theoretical developments in probabilistic inference. Next, I will move on to describe practical applications of deep learning technology, providing new approaches in psychology, neuroscience and neurotechnology. Murphy’s law permitting I will end with a brief demonstration of our most recent work.
*Abstract Taco Cohen*
In this talk I will give an overview of work being done on group equivariant convolutional networks (G-CNNs). Such networks incorporate prior knowledge about invariances in the data, resulting in substantial improvements in data efficiency and speed of convergence. I will survey various applications, such as analysis of satelite imagery, classification and segmentation of medical images such as CT scans and diffusion tensor images with 2D and 3D G-CNNs, molecular energy regression with Spherical CNNs, and others. Then I will present a high-level overview of a general field theory of equivariant networks that is now emerging, which relates these networks to fundamental ideas in mathematics and physics, and allows us to systematically classify existing architectures as well as guide the development of new ones.