Fwd: [Brown CS Talks] BigAl talk: Richard Sutton in Room 368 on 8/14/2024 at 1:00 PM

[Ricardo Fabbri]

Olá,

Nao sei se da pra entrar no Link sendo de fora,

mas a leitura do abstract abaixo já é levemente interessante, mesmo que nao tao novo o conceito.

Abraço,
--
Ricardo Fabbri
Professor of Computer Engineering

IPRJ/UERJ Nova Friburgo Graduate Program – CAPES 6
rfabbri.github.io

---------- Forwarded message ---------
From: <rece...@cs.brown.edu>
Date: Fri, 26 Jul 2024 at 09:47
Subject: [Brown CS Talks] BigAl talk: Richard Sutton in Room 368 on 8/14/2024 at 1:00 PM
To: <ta...@lists.cs.brown.edu>

Richard Sutton

Professor, University of Alberta and Research Scientist, Keen Technologies
Wednesday, August 14, 2024 at 1:00 PM

Room 368 (CIT - 3rd floor)

Zoom link: https://brown.zoom.us/j/99250354332
Add to calendar

BigAI Talk: Dynamic Deep Learning

Abstract: Despite great successes, current deep learning methods cannot learn effectively during normal operation, which makes them ill-suited for reinforcement learning or, really, for any general intelligence. In particular, conventional artificial neural networks fail catastrophically in classic supervised learning testbeds, such as ImageNet, when those testbeds are extended to require ongoing learning. In this talk, I argue that this failure is not inherent in neural networks, but just of the algorithms currently used. For example, a simple modification of the standard backpropagation algorithm, known as continual backpropagation, greatly improves performance in continual learning settings. Such results suggest exploring network learning algorithms explicitly designed for continual and reversible change, such as Dynamic deep learning networks, which continually adapt at multiple levels including 1) their weights, 2) their step-size parameters, and 3) their interconnection structure.

Richard S. Sutton is research scientist at Keen Technologies, professor in the Department of Computing Science at the University of Alberta, chief scientific advisor of the Alberta Machine Intelligence Institute (Amii), and fellow of the Royal Society of London, the Royal Society of Canada, the Association for the Advancement of Artificial Intelligence, Amii, and CIFAR. He received a PhD in computer science from the University of Massachusetts in 1984 and a BA in psychology from Stanford University in 1978. Prior to joining the University of Alberta in 2003, he worked in industry at AT&T Labs and GTE Labs, and in academia at the University of Massachusetts. He helped found DeepMind Alberta in 2017 and worked there until its dissolution in 2023. At the University of Alberta, Sutton founded the Reinforcement Learning and Artificial Intelligence Lab, which now consists of ten principal investigators and about 100 people altogether. Sutton is co-author of the textbook Reinforcement Learning: An Introduction, and his scientific publications have been cited more than 140,000 times. He is also a libertarian, a chess player, and a cancer survivor.

Host: Professor George Konidaris

 

_______________________________________________
The Computer Science Department is located at 115 Waterman St., Providence, RI. Our Calendar of Events at http://www.cs.brown.edu/events lists all of our talks.

If you would like to be removed from (or added to) our seminar mailing list, please send email containing your request to mailto:talks...@lists.cs.brown.edu.

We'd very much appreciate your forwarding these announcements to colleagues who may also have an interest. Thanks for your cooperation.