Learning Data Representation: Hierarchies and Invariance


Learning Data Representation: Hierarchies and Invariance

November 22-24, 2013 | McGovern Institute for Brain Research, MIT

The goal of the meeting is to investigate advances and challenges in learning "good representations" from data, in particular representations that can reduce the complexity of later supervised learning stages. The meeting will gather experts in the field to discuss current and future challenges for the theory and applications of learning representations.

(new!) Videos and slides of the talks and panels are now available through CBMM and MIT TechTV (links included in the agenda/schedule)

The workshop is organized by Tomaso Poggio and Lorenzo Rosasco and sponsored by the Laboratory for Computational and Statistical Learning (LCSL), joint between the Istituto Italiano di Tecnologia and Massachusetts Institute of Technology and the Center for Brains, Minds and Machines (CBMM).


  • Pierre Baldi (University of California, Irvine)
  • Mikhail Belkin (Ohio State University)
  • Joachim M. Buhmann (ETH, Brown University)
  • Charles Cadieu (Massachusetts Institute of Technology)
  • Dorin Comaniciu (Siemens Corporate Technology)
  • James DiCarlo (Massachusetts Institute of Technology)
  • Susan Epstein (Hunter College)
  • William Freeman (Massachusetts Institute of Technology)
  • Boris Katz (Massachusetts Institute of Technology)
  • Gabriel Kreiman (Harvard University)
  • Yann LeCun (New York University)
  • Joel Z. Leibo (DeepMind)
  • Lakshminarayanan Mahadevan (Harvard University)
  • Stéphane Mallat (Ecole Normale Superieur)
  • Giorgio Metta (Istituto Italiano di Tecnologia)
  • Pablo A. Parrilo (Massachusetts Institute of Technology)
  • Pietro Perona (California Institute of Technology)
  • Nicolas Pinto (Massachusetts Institute of Technology)
  • Tomaso Poggio (Massachusetts Institute of Technology)
  • Ben Recht (University of California, Berkeley)
  • Maximilian Riesenhuber (Georgetown University)
  • Lorenzo Rosasco (Università di Genova, Istituto Italiano di Tecnologia)
  • Ruslan Salakhutdinov (University of Toronto)
  • Giulio Sandini (Istituto Italiano di Tecnologia)
  • Guillermo Sapiro (Duke University)
  • Thomas Serre (Brown University)
  • Jean-Jacques Slotine (Massachusetts Institute of Technology)
  • Stefano Soatto (University of California, Los Angeles)
  • Charles Stevens (Salk Institute)
  • Josh Tenenbaum (Massachusetts Institute of Technology)
  • Carlo Tomasi (Duke University)
  • Shimon Ullman (Weizmann Institute of Science)
  • Leslie Valiant (Harvard University)
  • Alessandro Verri (Università di Genova)
  • Ramesh Visvanathan (Siemens, Frankfurt Institute for Advanced Studies/Goethe University)
  • Alan Yuille (University of California, Los Angeles)

Workshop Agenda

There will be four sessions, each one with a set of talks and a panel discussion. 

  • Session 1: Early Features in Vision
  • Session 2: Learning Features and Representations
  • Session 3: Learning Invariances and Hierarchies
  • Session 4: Beyond Feedforward Architectures

Schedule: pdf
Videos/slides: Use the links below for the presentation slides (kindly made available by all participants) and browse the videos listed to the right.

Friday November 22, 2013

MIT Bldg 46-3002, Singleton Auditorium

14:30 Introduction to the workshop | T. Poggio (video), L. Rosasco (slides)

Session: Early Features in Vision | Chair: C. Tomasi

14:50 Filters and other potions for early vision and recognition | P. Perona (slides)
15:30 Transformations in early vision from a symmetry argument | C. Stevens (slides)
16:10 Break
16:40 Hierarchical models of the visual cortex | T. Serre (slides)
17:20 Panel discussion | C. Tomasi [moderator] (slides), W. Freeman (slides), G. Sandini (slides) + Speakers
8:00 Reception at "Catalyst Restaurant"

Saturday November 23, 2013

MIT Bldg 46-3002, Singleton Auditorium

  8:30 am Breakfast

Session: Learning Features and Representations | Chair: A. Verri

  9:30 Latent structure beyond sparse codes | B. Recht (slides)
10:10 Dictionary learning: 3+ snippets on learning features | G. Sapiro (slides)
10:50 Break
11:20 Panel discussion | A.Verri [moderator], L. Rosasco (slides), M. Belkin (slides), J. Buhmann (slides) + Speakers
12:20 Lunch

Session: Learning Invariances and Hierarchies | Chair: J. DiCarlo

14:00 Learning invariant feature hierarchies | Y. LeCun (slides)
14:50 Scattering bricks to build invariants for perception | S. Mallat (slides)
15:40 Break
16:10 M-theory | T. Poggio (slides)
17:00 Panel discussion | J. DiCarlo [moderator] (slides), P. Baldi (slides), M. Riesenhuber + Speakers
18:30 Dinner (up to individual participants)

Sunday November 24, 2013

MIT Bldg 46-3002, Singleton Auditorium

 8:30 am Breakfast

Session: Beyond Feedforward Architectures | Chair: J. Tenenbaum

  9:00 The role of mobility and control in the inference of representations | S. Soatto (slides)
  9:40 Atoms of recognition | S. Ullman (slides)
10:20 Break
10:50 Compositional models: complexity of representation and inference | A. Yuille (slides)
11:30 Panel discussion | J. Tenenbaum [moderator] (slides), R. Salakhutdinov (slides) + Speakers

12:30 Final remarks | T. Poggio (slides)