Dr. Roland Vollgraf: "Learning Set-equivariant Functions with SWARM Mappings"

27. Jun

Dr. Roland Vollgraf: "Learning Set-equivariant Functions with SWARM Mappings"

Im Rah­men des For­schungs­kol­lo­qui­ums Wirt­schafts­in­for­ma­tik und Data Sci­ence re­fe­riert Dr. Roland Vollgraf von Zalando über "Learning Set-equivariant Functions with SWARM Mappings"


Da­tum und Ort:  27. Juni 2019    12.15 Uhr   C 14.203


Deep Neural Networks and their ability to model arbitrarily complex functions have revolutionised all areas of Machine Learning and Artificial Intelligence. The special domain of set-equivariant functions since recently enjoys growing interest in the research community. Set-equivariant functions are functions from sets to sets of the same cardinality, such that the ordering of the entities (the set elements) in the population (the sets) doesn’t matter and any permutation at the input results in the same permutation at the output.I will introduce a novel deep learning architecture that implements such functions and allows them to be learned efficiently from data. The architecture is based on a gated recurrent neural network which iteratively carries on all entities, at the same time syncing with the progression of the whole population. In reminiscence to this patterns, which can be frequently observed in nature, we call our approach SWARM mapping.