ML papers keep saying that abstraction is compression and then fall in love with information-theoretical frameworks. Please repeat after me: a) abstraction is not lossless compression, b) there is no universal compression/representation. 1/9 Why are there no universals? Deep down, because of the same reasons, specific biological systems do not only come from physics from scratch. An ecosystem might discover one way of many of doing things, and then the rest specializes in exploiting that accident. 2/9

Let’s talk abstractions and symmetries for when there are multiple ways of doing something. They are related to AI, math and life. Symmetries introduce what Jorge Luis Borges called “apparent differences.” The typical solution is to represent in a way that ignores part of that. Symmetries are both enemies and allies for AI. It’s hard to know when an input should be considered the same as another. If anything. Deep Learning is good at detecting similarities ignoring unimportant variations. E.g. image classification can ignore colour variations.

Reinforcement Learning and Planning? Submissions are welcome to the workshop "Bridging the Gap Between AI Planning and Reinforcement Learning (PRL)." Deadline Feb 24. Workshop date: June 8 or 9 (TBD). prl-theworkshop.github.io #RL #AI #Planning #ML #Reasoning #icaps #prl2021 /1 In the last edition, we accepted 20+ papers, had 5 invited speakers, 4 discussions and 100+ Zoom participants. See papers, posters and talks recording at icaps20subpages.icaps-conference.org/workshops/prl/ /2