Deep Learning Methods for Vision

CVPR 2012 Tutorial

9:00am-5:30pm, Sunday June 17th, Ballroom D (Full day)

Rob Fergus (NYU),
Honglak Lee (Michigan),
Marc'Aurelio Ranzato (Google)
Ruslan Salakhutdinov (Toronto),
Graham Taylor (Guelph),
Kai Yu (Baidu)


Hand-designed features such as SIFT and HOG underpin many successful object recognition approaches. However, these only capture low-level edge information and it has proven difficult to design features that effectively capture mid-level cues (e.g. edge intersections) or high-level representation (e.g. object parts). However, recent developments in machine learning, known as "Deep Learning", have shown how hierarchies of features can be learned in an unsupervised manner directly from data. This tutorial will describe these feature learning approaches, as applied to images and video.

The tutorial will start by motivating the need to learn features, rather than hand-craft them. It will then introduce several basic architectures, explaining how they learn features, and showing how they can be "stacked" into hierarchies that can extract multiple layers of representation. Throughout, links will be drawn between these methods and existing approaches to recognition, particularly those involving hierarchical representations. The final part of the lecture will examine the current performances obtained by feature learning approaches on a range of standard vision benchmarks, highlighting their strengths and weaknesses.


9:00am Introduction PPT (Fergus) [1h]
10:00am Coffee Break [30m]
10:30am Sparse Coding PPT (Yu) [1h]
11:30am Neural Networks PDF Code (Ranzato) [1h]
12:30pm Lunch [1h]
1:30pm Restricted Boltzmann Machines PDF (Lee) [1h]
2:30pm Deep Boltzmann Machines PDF (Salakhutdinov) [30m]
3:00pm Coffee Break [30m]
3:30pm Transfer Learning PDF (Salakhutdinov) [30m]
4:00pm Motion and Video PDF (Taylor) [1h]
5:00pm Summary / Q & A [30m]

Speaker Biographies

Rob Fergus

Rob Fergus is an Assistant Professor of Computer Science at the Courant Institute of Mathematical Sciences, New York University. He received a Masters in Electrical Engineering with Prof. Pietro Perona at Caltech, before completing a PhD with Prof. Andrew Zisserman at the University of Oxford in 2005. Before coming to NYU, he spent two years as a post-doc in the Computer Science and Artificial Intelligence Lab (CSAIL) at MIT, working with Prof. William Freeman. He has received several awards including a CVPR best paper prize (2003), a Sloan Fellowship (2011) and an NSF Career award (2012).
Honglak Lee

Honglak Lee is currently an Assistant Professor of Computer Science at the University of Michigan, Ann Arbor. He recevied his PhD from Stanford Unviersity, advised by Andrew Ng. His research interests lie in machine learning and its application to a range of perception problems in the fields of artificial intelligence, such as computer vision, robotics, audio recognition, and text processing.
Marc'Aurelio Ranzato

Marc'Aurelio Ranzato is currently a Research Scientist at Google. Before joining Google in the fall 2011, he was a post-doctoral fellow in Machine Learning, University of Toronto, working with Geoffrey Hinton. He did his Ph.D. in Computer Science at New York University in Yann LeCun's group. His interestes include Machine Learning, Computer Vision and, more generally, Artificial Intelligence. He has worked on unsupervised learning algorithms, in particular, hierarchical models and deep networks.
Ruslan Salakhutdinov

Ruslan Salakhutdinov received his PhD in machine learning from the University of Toronto in 2009. After spending two post-doctoral years at the Massachusetts Institute of Technology Artificial Intelligence Lab, he joined the University of Toronto as an Assistant Professor in the Departments of Statistics and Computer Science. Dr. Salakhutdinov's primary interests lie in statistical machine learning, Bayesian statistics, probabilistic graphical models, and large-scale optimization. He is the recipient of the NSERC Postdoctoral Fellowship, Canada Graduate Scholarship, and a Scholar of the Canadian Institute for Advanced Research.
Graham Taylor

Graham Taylor recently joined University of Guelph as an Assistant Professor of Engineering. He was previously a postdoc at NYU, working with Chris Bregler, Rob Fergus, and Yann LeCun. He completed his PhD at the University of Toronto in 2009, co-advised by Geoffrey Hinton and Sam Roweis. His interests are in statistical machine learning and biologically-inspired computer vision, with an emphasis on unsupervised learning and time series analysis. Much of his work studies human movement.

Kai Yu

Kai Yu recently jointed Baidu as Director of Multimedia Department, in charge of search technologies and products involving video, speech and music. Previously, he was head of the Media Analytics Department of NEC Labs in Silicon Valley, California, leading the development of intelligent systems for machine learning, image recognition, multimedia search, video surveillance, recommendation, data mining, and human-computer interface. He obtained PhD in Computer Science at University of Munich, Germany.


This work was partially supported by the National Science Foundation Career Award #1149633. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.