Alison Gopnik - Babies, brains and Bayes (Banquet speaker)
The banquet will be Thursday August 16th at the Casino Ballroom (moved from the Descanso beach club)
How do young children learn so much about the world so quickly and accurately? Many researchers have proposed that children implicitly formulate structured hypotheses about the world and then use evidence to test and revise those hypotheses. I’ll describe extensive research has shown over the past ten years that even two-year-olds formulate causal hypotheses and test and evaluate them against the data in a normatively accurate way. But this work raises several problems. Where do those hypotheses come from? What algorithms could children implicitly use to approximate ideal Bayesian inference? How can we reconcile children’s striking inferential success with their apparent variability and irrationality? And are there developmental differences in the ways that children and adults learn?
I will suggest that all these questions can be illuminated by thinking about children’s hypothesis generation as a sampling process – an idea with a long and successful history in computer science. Preschoolers may use sampling to generate hypotheses, and this may explain both their successful inferences and the variability in their behavior. In fact, in recent research we have discovered that children’s causal learning has some of the signatures of sampling. In particular, the variability in children’s responses reflects the probability of their hypotheses. Moreover, we have shown that children may use a particular type of sampling algorithm. We also found that preschoolers were actually more open-minded learners than older children and adults in some tasks. They seemed more likely to accept a wide range of novel hypotheses. This suggests that they may search at a “higher temperature” than adults do. From an evolutionary perspective the transition from childhood to adulthood may be nature’s way of performing simulated annealing.
Alison Gopnik is a professor of psychology and affiliate professor of philosophy at the University of California at Berkeley. She received her BA from McGill University and her PhD. from Oxford University. She is an internationally recognized leader in the study of children’s learning and development and was the first to argue that children’s minds could help us understand deep philosophical questions. She was also one of the founders of the field of “theory of mind”, and more recently introduced the idea that probabilistic models and Bayesian inference could be applied to children’s learning. She is the author or coauthor of over 100 journal articles and several books including “Words, thoughts and theories” MIT Press, 1997, and the bestselling and critically acclaimed popular books “The Scientist in the Crib” William Morrow, 1999, and “The Philosophical Baby; What children’s minds tell us about love, truth and the meaning of life” Farrar, Strauss and Giroux, 2009. She has also written widely about cognitive science and psychology for Science, The New York Times, Scientific American, The Times Literary Supplement, The New York Review of Books, New Scientist and Slate, among others. And she has frequently appeared on TV and radio including “The Charlie Rose Show” and “The Colbert Report”. She has three sons and lives in Berkeley, California with her husband Alvy Ray Smith.
Tom Mitchell - Never ending learning (Cancelled)Carnegie Mellon University
This talk describes our research to build a Never-Ending Language Learner (NELL) that runs 24 hours per day, forever, learning to read the web. Each day NELL extracts (reads) more facts from the web, and integrates these into its growing knowledge base of beliefs. Each day NELL also learns to read better than yesterday, enabling it to go back to the text it read yesterday, and extract more facts, more accurately. NELL has been running 24 hours/day for over two years now. The result so far is a collection of 15 million interconnected beliefs (e.g., servedWtih(coffee, applePie), isA(applePie, bakedGood)), that NELL is considering at different levels of confidence, along with hundreds of thousands of learned phrasings, morphoogical features, and web page structures that NELL uses to extract beliefs from the web. Track NELL's progress at http://rtw.ml.cmu.edu.
Tom M. Mitchell founded and chairs the Machine Learning Department at Carnegie Mellon University, where he is the E. Fredkin University Professor. His research uses machine learning to develop computers that are learning to read the web, and uses brain imaging to study how the human brain understands what it reads. Mitchell is a member of the U.S. National Academy of Engineering, a Fellow of the American Association for the Advancement of Science (AAAS), and a Fellow of the Association for the Advancement of Artificial Intelligence (AAAI). He believes the field of machine learning will be the fastest growing branch of computer science during the 21st century. Mitchell's web page is http://www.cs.cmu.edu/~tom.
Bill Macready - Quantum computing meets machine learningD-Wave systems
Quantum Computing offers the theoretical promise of dramatically faster computation through direct utilization of the underlying quantum aspects of reality. This idea, first proposed in the early 1980s, exploded in interest in 1994 with Peter Shor's discovery of a polynomial time integer factoring algorithm. Today the first experimental platforms realizing small-scale quantum algorithms are becoming commonplace. Interestingly, machine learning may be the "killer app" for quantum computing. We will introduce quantum algorithms, with focus on a recent quantum computational model that will be familiar to researchers with a background in graphical models. We will show how a particular quantum algorithm -- quantum annealing -- running on current quantum hardware can be applied to certain optimization problems arising in machine learning. In turn, we will describe a number of challenges to further progress in quantum computing, and suggest that machine learning researchers may be well-positioned to drive the first real-world applications of quantum annealing.
William Macready was trained as a theoretical physicist, but has spent most of his career building optimization and machine learning based systems deployed in commercial applications. He currently leads the algorithm group at D-Wave Systems, a group charged with developing algorithms and applications running on D-Wave's worlds-first large scale quantum annealing hardware. His team has implemented many of the largest scale quantum algorithms to date, including integer factoring, Ramsey number determination, binary and structured classification, and L0 penalized unsupervised learning.
Pedro Felzenszwalb - Graphical models for computer visionBrown University
Graphical models provide a powerful framework for expressing and solving a variety of inference problems. The approach has had an enormous impact in computer vision. In this talk I will review some of the developments that have enabled this impact, focusing on efficient algorithms that exploit the structure of vision problems. I will discuss several applications including the low-level vision problem of image restoration, the mid-level problem of segmentation and the high-level problem of model-based recognition. I will also discuss some of the current challenges in the area.
Pedro F. Felzenszwalb is an Associate Professor at Brown University. He received his PhD from MIT in 2003. His main research interests are in computer vision, artificial intelligence and discrete algorithms. He was awarded the IEEE CVPR Longuet-Higgins Prize for a contribution that stood the test of time in 2010. He also received the PASCAL Visual Object Challenge Lifetime Achievement prize in 2010 for his work on visual object detection. Webpage: http://www.cs.brown.edu/~pff
Alexander Gray - Machine Learning on (Astronomically) Large DatasetsGeorgia Institute of Technology and Skytree Inc.
There is much talk of "big data" these days, surrounding a seismic shift currently underway in industry. Much of what is being discussed today is reminiscent of, and perhaps can be informed by, efforts beginning at least two decades ago in astronomy -- which was forced into that pioneering position by some unique scientific constraints and motivating problems that we'll discuss. So what are the best ideas that have been developed over the years for doing machine learning on massive datasets? Toward scaling up the most popular textbook machine learning methods across seven basic types of statistical tasks, we'll first identify seven main types of computational bottlenecks. Then we'll consider seven cross-cutting classes of computational techniques which characterize the current fastest algorithms for these bottlenecks, discussing their strengths and weaknesses. We'll end with some exhortations regarding where more foundational research is needed.
Alexander Gray received bachelor's degrees in Applied Mathematics and Computer Science from the University of California, Berkeley and a PhD in Computer Science from Carnegie Mellon University, and is an Associate Professor in the College of Computing at Georgia Tech heading up its new Center for Big Data Analytics and Machine Learning, and CTO of Skytree Inc., "the machine learning company". His research group, the FASTlab, aims to algorithmically scale up all of the major practical methods of machine learning to massive datasets and has developed a number of fast algorithms for several key problems, as well as new statistical methodology. He began working with massive astronomical datasets in 1993 (long before the current fashionable talk of ³big data²) at NASA's Jet Propulsion Laboratory in its Machine Learning Systems Group. High-profile applications of his large-scale ML algorithms have been described in staff written articles in Science and Nature, including contributions to work selected by Science as the Top Scientific Breakthrough of 2003. His work has won or been nominated for seven best paper awards and he is a National Science Foundation CAREER Award recipient, National Academy of Sciences Kavli Scholar, and frequent advisor on the topic of big-data machine learning at research conferences, government agencies, and corporations, recently serving on the National Academy of Sciences Committee on the Analysis of Massive Data.
Judea Pearl - The Latent Powers of the Do-CalculusUniversity of California, Los Angeles
The do-calculus was developed in 1995 to facilitate the identification of causal effects in non-parametric models. The completeness proofs of Huang and Valtorta (2006) and Shpitser and Pearl (2006) have put this identification problem to rest. Recent explorations unveil the usefulness of the do-calculus in three additional areas: mediation analysis (Pearl 2012), transportability (Pearl and Bareinboim 2011) and meta-synthesis. Meta-synthesis (freshly coined) is the task of fusing empirical results from several diverse studies, conducted on heterogeneous populations and under different conditions, so as to synthesize a causal relation in some target environment, potentially different from those under study. The talk will survey these results with emphasis on the challenges posed by meta-synthesis.
For background material, see http://bayes.cs.ucla.edu/csl_papers.html