Final Schedule

December 5, 2010

18:15Meet at Hyatt Regency Vancouver lobby to walk to restaurant.
19:00 Pre-workshop dinner at location:
Downtown Kirin Restaurant

1172 Alberni Street
Telephone: (604) 682-8833

December 6, 2009 - Hyatt Regency Vancouver, Georgia A room

07:30Registration & poster setup
08:00Breakfast
08:45Welcome Presentation Slides
09:00 Invited Speaker Sally Goldman (Google Research/Washington University in St. Louis)
Self-Pruning Prediction Trees
We present a generalization of decision trees for performing real-valued predictions. In our model each node of the tree is associated with a confidence value, and a prediction is formed by summing the confidence values along the path from the root to a leaf node. We define a complexity measure for prediction trees using a variation norm of their outputs. In the context of prediction trees, this variation norm amounts to p-norms of the confidence-rated partitions induced by the decision tree. We then cast the problem of learning a prediction tree as a penalized empirical risk minimization task. By choosing sparsity promoting norms, we are able to derive a tree growing procedure, which terminates automatically. The end result is a learning algorithm for decision and prediction trees, in which growing and pruning take place in tandem and are tightly coupled. Our approach is applicable to various loss functions and different variation norms. We close with some preliminary experiments comparing our approach to CART, a standard tree growing and pruning procedure. This work is joint with Yoram Singer and John Duchi.

09:35Oral Session I
Ilknur Icke (City University of New York, The Graduate Center)
Multi-Objective Genetic Programming Projection Pursuit for Exploratory Data Modeling
Presentation Slides
Talya Meltzer (The Hebrew University of Jerusalem)
Convergent message passing algorithms - a unifying view
Presentation Slides
Dorota Glowacka (University College London)
Sifting through Images with Multinomial Relevance Feedback
Presentation Slides
10:20Coffee break
10:30Invited Speaker Ming Hua (Facebook)
Machine Learning Challenges in Facebook
Being the world's largest social network, Facebook provides various social utilities for more than 500 million users. The people, the friend connections, and the objects people interact with form a large dynamic social graph with rich information. Massive content is being generated on the large social graph. The scale of the social graph and the variety of user generated content pose grand machine learning challenges in various services on Facebook, such as news feed, search, advertising and site integrity. This talk will discuss some interesting machine learning problems we encounter in Facebook and share the insights we learnt.

11:05 Poster Session I and parallel Small Group Discussion I:
first names A-L
12:05 Lunch
13:05Invited Speaker Isabelle Guyon (ClopiNet) Presentation Slides
Competitions in machine learning: the fun, the art, and the science
Challenges have recently proved a great stimulus for research in machine learning, pattern recognition, and robotics. Robotics contests seem to be particularly popular, the most visible ones probably being the DARPA grand challenges of autonomous ground vehicle navigation and RoboCup featuring several challenges for robots including playing soccer or rescuing people. The European network of excellence PASCAL has actively sponsored a number of challenges around hot themes in machine learning, which have punctuated workshops at NIPS and other conferences. These contests are oriented towards scientific research and the main reward for the winners is to disseminate the product of their research and obtain recognition. In that respect, they play a different role than challenges like the Netflix prize, which offer large monetary rewards for solving a task of value to the Industry (movie referral in than particular case), but are narrower scope. Attracting hundreds of participants and the attention of a broad audience of specialists as well as sometimes the general public, these events have been important in several respects: (1) pushing the state-of-the art, (2) identifying techniques which really work, (3) attracting new researchers, (4) raising the standards of research, (5) giving the opportunity to non-established researchers to make themselves rapidly known. The good news is that women have understood that opportunity and are well represented among the top ranking participants (see http://clopinet.com/challenges/people.html). Since 2003, we have been organizing challenges in machine learning, including 2 for NIPS. We addressed problems of both fundamental and practical interest in machine learning, data mining or statistics, illustrated with data from various domains. For instance, in 2003 we organized a challenge on feature selection, in 2006 and 2007, we organized a series of challenges on model selection, and in 2008 two challenges on causality. Our challenge platforms, which remain open for post-challenge submissions, are constantly in use by students and have been used in practical work in our own classes and those of other professors throughout the world. We take great care of giving to the participants opportunities publish in reputable conferences proceedings or journals like JMLR. In 2009, we organized the KDD cup (the oldest data mining competition). We chose a marketing problem with the goal of identifying data mining techniques capable of rapidly building predictive models and scoring new entries on a large customer database from the French Telecom company Orange. The challenge attracted over 450 participants from 46 countries. We attribute its popularity to several factors: (1) A generic problem relevant to the Industry, but presenting a number of scientific and technical challenges. (2) Prizes (Orange offers 10000 Euros in prizes). (3) A well designed protocol and web site (we benefitted from past experience). (4) An effective advertising campaign using mailings and a teleconference to answer potential participants questions.

13:40Oral Session II
Pooja Viswanathan (University of British Columbia)
Adaptive Prompting for Intelligent Wheelchairs
Presentation Slides
Rasna Walia (Iowa State University)
Improving the State of the Art in Machine Learning Methods for Protein-RNA Interface Prediction
Presentation Slides
Suchi Saria (Stanford University)
Decoding longitudinal Electronic Health Record (EHR) data
Presentation Slides
14:25Coffee break
14:35Poster session II and parallel Small Group Discusssion II:
first name M-Z
15:35Invited Speaker Raquel Urtasun (TTI Chicago)
3D Urban Scene Understanding from Movable Platforms
3D scene understanding is key for the success of applications such as autonomous driving and robot navigation. However, existing approaches either produce a mild level of understanding, e.g., segmentation, object detection, or are not accurate enough for these applications, e.g., 3D pop-ups. In this talk I will show a generative model of 3D urban scenes that takes into account dependencies between static and dynamic features, and is able to infer the geometric (e.g., street orientation) and topological (e.g., number of intersecting streets) properties of the scene layout, as well as the semantic activities occurring in the scene, e.g., traffic situations at an intersection. Furthermore, I will show that this global level of understanding provides the context necessary to disambiguate current state-of-the-art detectors. This is joint work with Andreas Geiger and Martin Lauer.
16:10Career Panel - Introduction and Group Mentoring
17:10Closing remarks & poster take-down
17:30End of Workshop