Are you smarter than a baby? If so, why?

M208, kl 12-13, 28.september

  • 28.9.2022, 12:00 - 13:00, Háskólinn í Reykjavík

On Wednesday 28th of September, we welcome Henry Minsky for a lecture at Reykjavík University. Question: How do humans develop intelligence so rapidly, going from a helpless newborn infant to a toddler with a high degree of 'common sense' in under two years? In that short time, they acquire hundreds of thousands or perhaps millions of pieces of carefully interconnected causal knowledge, organized to use in goal-oriented behavior to reliably recognize and manipulate objects in their world. The acquired knowledge appears to be automatically organized into hierarchical models of their environment for planning, and reasoning.

While there has been great recent progress in neural-network reinforcement learning, it provides only a piece of the puzzle of how to build common-sense reasoning. A fundamental question in RL remains "who is giving the system a reward, and for what?". Humans are capable of a high degree of self-supervised learning through play, while equally capable of learning by imitation and observation of others. Reinforcement learning alone does not exhibit the rapid explosion in skill acquisition that humans do, while requiring orders of magnitude more training data. We are developing a highly imitative self-supervised learning agent system, called Leela.

The learning algorithms we use are guided by observations from human infant developmental psychology, and can learn tasks using polynomially fewer training trials than pure reinforcement learning techniques. The algorithms Leela uses are designed with an approach of "fast, cheap, and good enough" concept formation, which is designed to emulate the rapid exponential increase in reasoning and planning ability that humans get from fast concept acquisition and knowledge reuse. BIO: Henry Minsky is CTO of Leela AI, Inc., a venture backed startup out of MIT developing common sense AI-based tools to understand 'what's going on' in complex industrial environments. He is the driving force behind the core Leela AI software: For PhD work at the MIT AI Lab, Mr. Minsky extended a constructivist learning system with a vision system incorporating image processing capability developed while at Xerox PARC. Minsky built high speed network routing chips at MIT under the DARPA Critical Problems In Very Large Computer Systems project. He developed distributed location-based services and a wearable sensor network architecture at NTT DoCoMo. Prior to Leela AI, he was an early employee at Google Nest Labs developing cloud services, analysis, and ML for millions of networked IoT devices. (SB, MSc MIT).



Vinsamlegast athugið að á viðburðum Háskólans í Reykjavík (HR) eru teknar ljósmyndir og myndbönd sem notuð eru í markaðsstarfi HR. Hægt er að nálgast frekari upplýsingar á ru.is eða með því að senda tölvupóst á netfangið: personuvernd@ru.is
//
Please note that at events hosted at Reykjavik University (RU), photographs and videos are taken which might be used for RU marketing purposes. Read more about this on our ru.is or send an e-mail: personuvernd@ru.is