Deep X: Deep Learning with Deep Knowledge

We argue that a labeled graph is an appropriate description of world state and world events on a cognitive abstraction level, representing facts as subject-predicate-object triples. A prominent and very successful example is the Google Knowledge Graph, representing on the order of 100B facts. Labeled graphs can be represented as adjacency tensors which can serve as inputs for prediction and decision-making, and from which tensor models can be derived to generalize to unseen facts. We show how these ideas can be used, together with deep recurrent networks, for clinical decision support by predicting orders and outcomes. We discuss potential links to the memory and perceptual systems of the human brain and propose quantum algorithms to solve high-dimensional tensor decomposition problems. In connection with deep learning, this can be the basis for many technical solutions requiring memory and perception and might bridge modern physics and modern Al.

Student event: Meet the speaker

We invite you to a student-only discussion-round with Prof. Dr. Volker Tresp before his Munich Physics Colloquium talk. Be curious and feel free to ask any question.
Monday, 2nd of July 2018, 16:00 h, Room H 522 (5th floor), Fakultät für Physik der LMU, Schellingstraße 4, München.