Michael S. Lew
Professor, Deep Learning
Computer Science Department (LIACS)
Leiden University, Netherlands
Michael S. Lew is a Professor of Deep Learning at Leiden University. He is also the co-head of the Computational Imaging and Deep Learning Research Cluster and co-director of the LIACS Media Lab. He has over 200 multimedia and deep learning publications of which one paper is in the top 10 most cited papers out of the 17,000 published in the history of the ACM SIG Multimedia. Another was the most cited paper in the high impact neural network journal, Neurocomputing from 2018-2020. He has served on the ACM SIGMM Executive Board and the ACM SIGMM Steering Committee. He was the founding chair of the steering and executive committees for the ACM CIVR and ACM ICMR conferences and the founding editor-in-chief of the International Journal of Multimedia Information Retrieval (Springer and ACM SIGMM).
Title: Advances and Paradigms in DeepLearning
Deep learning approaches have become an important pillar for artificial intelligence and machine learning where major strides have been made in numerous well known grand challenge areas such as computer vision, object recognition, person identification, automatic driving, robotics, scientific imagery analysis and computer aided health diagnosis. In the first generation of CNN based deep learning methods, the primary goal was toward solving problems that had static conditions (e.g. single modality, fixed number of classes, whole image) and where very large training datasets were available. In the second generation, the scientists are relaxing those conditions and tackling problems such as small training datasets and incremental learning and forgetting. In this talk I introduce and discuss how we are taking early steps toward a better understanding of the fundamentals of deep learning approaches, proposing new neural architectures and tackling well known vision problems such as automatic image segmentation, object detection/recognition, combining multiple modalities using embedded spaces and learning incremental neural models where the number of classes can change.
The talk will be delivered online
School of Mathematical and Computer Sciences,
Heriot-Watt University, Edinburgh
Michael Lones is an Associate Professor of Computer Science at Heriot-Watt University in Edinburgh, Scotland. He received his PhD from the University of York in 2003, and has spent the last 20 years or so working in the areas of machine learning and data science, authoring around 80 publications. His early research focused on evolutionary algorithms and biomedicine, but his interests have since broadened out to include other methodological and application areas, including neural networks, complex systems, robotics, and computer security. He is on the editorial boards for the journals BioSystems and Genetic Programming and Evolvable Machines, and is a Senior Member of the IEEE and a member of the IEEE Technical Committee on Bioinformatics and Bioengineering.
IAPR Invited Speaker
Title: Evolutionary algorithms and neural networks: a match made in heaven?
Evolutionary algorithms and neural networks both model processes that are fundamental to biological systems, and both are considered to be pillars of computational intelligence. Reflecting their biological origins, these two approaches are largely complementary, and there are many examples of successfully melding the two. In this talk, I’m going to focus on recent advances in evolutionary algorithms, and try to convince you why they are worth considering as either an adjunct or alternative to off-the-shelf neural network-based approaches.
The talk will be delivered in person.