MaLGa Seminar: Tomaso Poggio, "Thoughts on today’s learning theory", February 4th 14:00 room 706/Youtube
1 view
Skip to first unread message
Matteo Santacesaria
Feb 2, 2022, 8:54:05 AM2/2/22
to malga-all, goa...@googlegroups.com
Apologies for multiple posting.
Abstract: I will describe a personal perspective on the present state of key problems in learning theory. Several different architectures that perform well have emerged, in addition to CNN, such as transformers, perceivers and MLP mixers. Is there a common motif to all of them and to their good performance? A natural conjecture is that these modern architecture are good for the approximation, learning and optimization of input-output mappings that can be represented by “sparse” functions that are effectively low-dimensional. In particular, these target functions are typically compositional functions with a function graph that has nodes each with dimensionality at most k, with k<<d where d is the dimensionality of the function domain.
Bio: Tomaso A. Poggio is the Eugene McDermott Professor in MIT’s Department of Brain and Cognitive Sciences and the director of the NSF Center for Brains, Minds and Machines at MIT. He is a founding member of the McGovern Institute as well as a member of the Computer Science and Artificial Intelligence Laboratory. A former Corporate Fellow of Thinking Machines Corporation, a former director of PHZ Capital Partners, Inc. and of Mobileye, he was involved in starting, or investing in several other high tech companies including Arris Pharmaceutical, nFX, Imagen, Digital Persona, Deep Mind and Orcam. He is one of the most cited computational scientists and has mentored PhD students and postdocs that are some of the today’s leaders in the science and engineering of intelligence.
--
Matteo Santacesaria
Assistant Professor
MaLGa - Machine Learning Genoa Center
Department of Mathematics
University of Genoa
Personal Homepage
We are pleased to announce the next MaLGa Seminar Series -
Statistical Learning and Optimization.
This event is part of the Ellis Genoa activities.
Speaker: Tomaso Poggio
Affiliation: Massachusetts
Institute of Technology
Date: Friday, February 4th, 2022
Time: 14:00 p.m.
Location: room 706, via Dodecaneso 35, Genova, IT
Live streaming will be available at 706DIMA - YouTube
Abstract: I will describe a personal perspective on the present state of key problems in learning theory. Several different architectures that perform well have emerged, in addition to CNN, such as transformers, perceivers and MLP mixers. Is there a common motif to all of them and to their good performance? A natural conjecture is that these modern architecture are good for the approximation, learning and optimization of input-output mappings that can be represented by “sparse” functions that are effectively low-dimensional. In particular, these target functions are typically compositional functions with a function graph that has nodes each with dimensionality at most k, with k<<d where d is the dimensionality of the function domain.
Bio: Tomaso A. Poggio is the Eugene McDermott Professor in MIT’s Department of Brain and Cognitive Sciences and the director of the NSF Center for Brains, Minds and Machines at MIT. He is a founding member of the McGovern Institute as well as a member of the Computer Science and Artificial Intelligence Laboratory. A former Corporate Fellow of Thinking Machines Corporation, a former director of PHZ Capital Partners, Inc. and of Mobileye, he was involved in starting, or investing in several other high tech companies including Arris Pharmaceutical, nFX, Imagen, Digital Persona, Deep Mind and Orcam. He is one of the most cited computational scientists and has mentored PhD students and postdocs that are some of the today’s leaders in the science and engineering of intelligence.
Matteo Santacesaria
Assistant Professor
MaLGa - Machine Learning Genoa Center
Department of Mathematics
University of Genoa
Personal Homepage
Reply all
Reply to author
Forward
0 new messages