Skip to main content Link Menu Expand (external link) Copy Copied
MoDL 2023
Splash photo of TTIC
MoDL Collaboration Meeting 2023
May 23-24, TTIC

Synopsis

Join us at TTIC from May 23rd-24th for talks and activities relevant to the Collaboration on the Theoretical Foundations of Deep Learning.

  • Scientific Program: beginning Tuesday, May 23rd at 11 AM CT, ending Wednesday, May 24th at 4 PM CT.
  • Full Program: The meeting will conclude with a picnic at The Point after the scientific program ends, on Wednesday, May 24th from 4 PM CT to 8 PM.

See the schedule below for the list of talks by internal members of the collaboration and external speakers. We have also planned a panel session, a “research speed dating” event, and a discussion session for research topics/open problems for students, postdocs, and faculty to enjoy.

Remote participation will be accommodated; see below.

Accommodations

Hotel accommodations will be at The Study Hotel, located in Hyde Park.

For any questions regarding travel accommodations, please email Brandie Jones

Schedule

Tuesday, May 23, 2023

  • 10:30 - 11:00 Welcome + Coffee
  • 11:00 - 11:45 Reza Gheissari “High-Dimensional Limit Theorems for Stochastic Gradient Descent”
  • 11:45 - 1:00 Poster Session + Lunch
  • 1:00 - 1:30 Nikhil Ghosh “A Universal Trade-off Between the Model Size, Test Loss, and Training Loss of Linear Predictors”
  • 1:35 - 2:05 Jingfeng Wu “Implicit Bias of Gradient Descent for Logistic Regression at the Edge of Stability”
  • 2:05 - 3:05 Research Speed Dating
  • 3:05 - 3:35 Break
  • 3:35 - 4:05 Lijia Zhou “Agnostic Interpolation Learning Beyond Linear Regression”
  • 4:05 - 4:35 Kangjie Zhou “Learning Time-Scales in Two-Layers Neural Networks”
  • 4:35 - 6:00 Breakout / Small Groups / Open Problems
  • 6:00 - 8:00 Cocktail Reception at TTIC

Wednesday, May 24, 2023

  • 9:45 - 10:00 Breakfast
  • 10:00 - 10:30 Promit Ghosal “Towards Understanding the Dynamics of Gaussian-Stein Variational Gradient Descent”
  • 10:30 - 10:40 Break
  • 10:40 - 11:30 Francesca Mignacco “Statistical Physics Insights into the Dynamics of Learning Algorithms”
  • 11:30 - 1:00 Lunch + Break
  • 1:00 - 2:00 Panel Discussion
  • 2:00 - 2:30 Parthe Pandit “Mechanism of Feature Learning in Deep Fully Connected Networks and Kernel Machines that Recursively Learn Features”
  • 2:30 - 2:45 Break
  • 2:45 - 3:30 Matus Telgarsky “Representational strengths and limitations of transformers”
  • 3:30 - 7:30 Picnic at The Point (Google Maps)

Reimbursements

TTIC will reimburse all travel expenses (airfare, train, bus, taxi, meals, etc.) related to this meeting. You must keep all of your receipts. For charges related to meals, an itemized receipt must be included, and a signed copy must be submitted in order to be processed.

  • For the purposes of business travel, domestic travel is defined as ‘travel to any destination within the United States, including Alaska and Hawaii, and all U.S. possessions, territories, and entities with free association status.” Travel to all other locations is considered foreign travel.
  • The allowable expense for domestic travel is a standard coach airfare or its equivalent. Domestic First Class or Business Class travel requires approval by the Chief Academic Officer and is allowed for the following reasons:
    • Physical disability
    • Lack of available space
    • Cancellation of service
    • Extraordinary situation
  • For more information regarding our reimbursement policy, please visit our Travel Policy.

After the meeting, an email with more information on how you submit your reimbursement will be sent out.

Organizers

Brandie Jones

TTIC

Table of contents