| Description |
An ICTP Virtual Meeting
The aim of the conference is to give young researchers the opportunity to present their results related to high-dimensional statistical problems, as arising in machine learning, inference or statistical physics.
Typical problems in high-dimensions include the detection and estimation of noisy signals (compressed sensing, PCA and tensor decomposition etc), the analysis of (deep) neural-networks and their learning and generalization dynamics, the detection of communities in large networks, or disordered spin systems. While these problems may seem of different natures, they actually share many similarities in their common phenomenology and the tools used to analyze them.
The core of the field is made of a very active, diverse and quickly expanding community of physicists, computer scientists, mathematicians, information theorists and engineers, with the common desire to tackle increasingly challenging problems at the forefront of data science. This conference aims at reinforcing the links among this interdisciplinary community, and in particular of its youngest theory-oriented members, and to bring forward the latest development happening in the high-dimensional world. Topics:
Participants interested to present a poster are kindly requested to submit their title and a short abstract in the application form. Speakers: L. BENIGNI, University of Chicago, USA M. CELENTANO, Stanford University, USA
L. CHIZAT, University of Paris-Saclay
SY. CHUNG, Massachusetts Institute of Technology, USA
E. DOBRIBAN, University of Pennsylvania, USA
Z. FAN, Yale University, USA
C. GERBELOT, ENS Paris, France
A. GLIELMO, SISSA, Italy
A. INGROSSO, ICTP, Italy
A. JACOT, EPFL, Switzerland
F. MASTROGIUSEPPE, University College London, UK
F. MIGNACCO, IPHT Saclay, France
T. MISIAKIEWICZ, Stanford University, USA
P. NAKKIRAN, Harvard University, USA
M. REFINETTI, ENS Paris, France
C. RUSH, Columbia University, USA
T. SCHRAMM, Stanford University, USA
S. VILLAR, Johns Hopkins University, USA
I. ZADIK, New York University, USA
J. ZARKA, ENS Paris, France
Registration: There is no registration fee. |
Booklet Poster Presenters
Youth in High-Dimensions | (smr 3602)
Go to day
-
-
16:25 - 16:30
Introductory words
Convener: Sebastian GOLDT (SISSA, Italy) -
16:30 - 18:00
#1 Statistical Physics and Theory of Deep Learning
Chair: Marylou GABRIÉ (NYU/Flatiron, USA)
-
16:30
Classifying High-dimensional Gaussian Mixtures: Where Kernel Methods Fail and Neural Networks Succeed
20'
Speaker: Maria REFINETTI (ENS Paris, France) Material: 
Video
-
16:50
The Deep Bootstrap Framework: A New Lens to Understand Generalization
20'
Speaker: Preetum NAKKIRAN (Harvard University, USA) Material:
Abstract
Video
-
17:10
Understanding the Dynamics of Multi-pass SGD in High Dimensional Non-convex Problems
20'
Speaker: Francesca MIGNACCO (IPHT Saclay, France) Material:
Abstract
Video
-
17:30
Questions & Answers
15'
Speaker: Maria REFINETTI (ENS, Paris), Preetum NAKKIRAN (Harvard University, USA), Francesca MIGNACCO (IPHT Saclay, France) - 17:45 Break 15'
-
16:30
Classifying High-dimensional Gaussian Mixtures: Where Kernel Methods Fail and Neural Networks Succeed
20'
-
18:00 - 19:15
#2 Feature Learning and Lazy Regimes in Neutral Networks
Chair: Marco MONDELLI (IST, Austria)
-
18:00
Minimum Complexity Interpolation in Random Features Models
20'
Speaker: Theodor MISIAKIEWICZ (Stanford University, USA) Material:
Abstract
Video
-
18:20
Implicit Regularization of Gradient Descent for Wide Two-layer Relu Neural Networks
20'
Speaker: Lénaïc CHIZAT (University of Paris-Saclay, France) Material:
Video
-
18:40
Saddle-to-Saddle Regime in Deep Linear Networks
20'
Speaker: Arthur JACOT (EPFL, Switzerland) Material:
Video
-
19:00
Questions & Answers
15'
Speaker: Theodor Jakub MISIAKIEWICZ (Stanford University, USA), Lénaïc CHIZAT (University of Paris-Saclay), Arthur JACOT (EPFL, Switzerland)
-
18:00
Minimum Complexity Interpolation in Random Features Models
20'
-
16:25 - 16:30
Introductory words
-
-
16:30 - 19:15
#3 Neural Networks and Data Analysis
Chair: Marylou GABRIÉ (NYU/Flatiron, USA)
-
16:30
Spectrum of Random Feed-Forward Neural Networks
20'
Speaker: Lucas BENIGNI (University of Chicago, USA) Material:
Video
-
16:50
Separation and Concentration in Deep Networks
20'
Speaker: John ZARKA (ENS Paris, France) Material:
Video
-
17:10
Ranking the Information Content of Distance Measures
20'
Speaker: Aldo GLIELMO (SISSA, Italy) Material:
Video
-
17:30
Questions & Answers
15'
Speaker: Luca BENIGNI (University of Chicago, USA), John ZARKA (ENS, Paris), Aldo GLIELMO (SISSA, Trieste, Italy) - 17:45 Break 15'
-
18:00
Discussions & Posters
1h15'
Gather Town
-
16:30
Spectrum of Random Feed-Forward Neural Networks
20'
-
16:30 - 19:15
#3 Neural Networks and Data Analysis
-
-
16:30 - 18:00
#4 Theoretical Neuroscience
Chair: Sebastian GOLDT (SISSA, Italy)
-
16:30
Evolution of Neural Activity in Circuits Bridging Sensory and Bbstract Knowledge
20'
Speaker: Francesca MASTROGIUSEPPE (University College London, UK) -
16:50
Optimal Learning with Excitatory and Inhibitory Synapses
20'
Speaker: Alessandro INGROSSO (ICTP, Italy) Material:
Video
-
17:10
Perceptual Manifolds in Biological and Artificial Neural Networks
20'
Speaker: SueYeon CHUNG (Columbia University, USA) -
17:30
Questions & Answers
15'
Speaker: Francesca MASTROGIUSEPPE (University College London, UK), Alessandro INGROSSO (ICTP, Trieste, Italy), Sueyeon CHUNG (Columbia University, USA) - 17:45 Break 15'
-
16:30
Evolution of Neural Activity in Circuits Bridging Sensory and Bbstract Knowledge
20'
-
18:00 - 19:35
#5 Statistics & Theoretical Computer Science
Chair: Marco MONDELLI (IST, Austria)
-
18:00
The All-or-Nothing Phenomenon: the Case of Sparse Tensor PCA
20'
Speaker: Ilias ZADIK (New York University, USA) Material:
Abstract
Video
-
18:20
Scalars are Universal: Gauge-equivariant Machine Learning, Structured Like Classical Physics
20'
Speaker: Soledad VILLAR (Johns Hopkins University, USA) Material:
Video
-
18:40
What Causes the Test Error? Going Beyond Bias-Variance via ANOVA
20'
Speaker: Edgar DOBRIBAN (University of Pennsylvania, USA) Material:
Abstract
Video
-
19:00
Statistical Query Algorithms and Low-Degree Tests Are Almost Equivalent
20'
Speaker: Tselil SCHRAMM (Stanford University, USA) -
19:20
Questions & Answers
15'
Speaker: Ilias ZADIK (New York University, USA), Soledad VILLAR (Johns Hopkins University, USA), Edgar DOBRIBAN (University of Pennsylvania, USA), Tselil SCHRAMM (Stanford University, USA)
-
18:00
The All-or-Nothing Phenomenon: the Case of Sparse Tensor PCA
20'
-
16:30 - 18:00
#4 Theoretical Neuroscience
-
-
16:30 - 19:30
#6 Approximate Message Passing & Computational Issues
Chair: Jean BARBIER (ICTP, Italy)
-
16:30
Beyond i.i.d. Gaussian Models: Exact Asymptotics with Realistic Data
20'
Speaker: Cedric GERBELOT (ENS Paris, France) Material:
Abstract
Slides
Video
-
16:50
Using AMP to Characterize the Optimal Type 1 Error and Power Trade-off for Sorted L1 Penalized Estimation (SLOPE)
20'
Speaker: Cynthia RUSH (Columbia University, USA) Material:
Abstract
Video
-
17:10
Approximate Message Passing Algorithms and Orthogonal Spin Glasses
20'
Speaker: Zhou FAN (Yale University, USA) Material:
Slides
Video
-
17:30
Debiasing the Lasso with Inaccurate Precision Matrix
20'
Speaker: Michael CELENTANO (Stanford University, USA) Material:
Abstract
Video
-
17:50
Questions & Answers
15'
Speaker: Cedric Daniel Rene GERBELOT-BARRILLON (Ecole Normale Superieure de Paris, France), Cynthia G.RUSH (Columbia University, USA), Zhou FAN (Yale University, USA), Michael CELENTANO (Stanford University, USA) -
18:05
Closing remarks
5'
Speaker: Jean BARBIER (ICTP, Italy) -
18:10
Discussions & Posters
1h20'
Gather Town
-
16:30
Beyond i.i.d. Gaussian Models: Exact Asymptotics with Realistic Data
20'
-
16:30 - 19:30
#6 Approximate Message Passing & Computational Issues