Simons Postdoctoral Fellow in the Theory of Computation Research Group
Optimization problems are ubiquitous in contemporary engineering. A principal barrier to solving several real-world optimization problems is input uncertainty. In this talk, I will present new tools to study probabilistic instances of integer programs. As an application, I will show a phase-transition phenomenon in a simple distribution model for random integer programs. Our main tool is an elementary connection between integer programming and matrix discrepancy. I will describe this connection and derive matching upper and lower bounds on the discrepancy of random Gaussian matrices.
Based on joint work with Santosh Vempala.
Karthekeyan Chandrasekaran is a Simons Postdoctoral Research Fellow at Harvard University. He obtained his B. Tech. in Computer Science and Engineering from the Indian Institute of Technology, Madras and his Ph.D. in Algorithms, Combinatorics, and Optimization from Georgia Tech. His primary research interests are in combinatorial optimization, probabilistic methods and analysis, and randomized algorithms.
Building E62 Map
MIT general map location link
Building 32 Map
University of California - Berkeley
Host: Regina Barzilay and Tommi Jaakkola
Host Affiliation: CSAIL
Clustering involves placing entities into mutually exclusive
categories. We wish to relax the requirement of mutual exclusivity,
allowing objects to belong simultaneously to multiple classes, a
formulation that we refer to as "feature allocation." The first step
is a theoretical one. In the case of clustering the class of
probability distributions over exchangeable partitions of a dataset
has been characterized (via exchangeable partition probability
functions and the Kingman paintbox). These characterizations support
an elegant nonparametric Bayesian framework for clustering in which
the number of clusters is not assumed to be known a priori. We
establish an analogous characterization for feature allocation; we
define notions of "exchangeable feature probability functions" and
"feature paintboxes" that lead to a Bayesian framework that does not
require the number of features to be fixed a priori. The second step
is a computational one. Rather than appealing to Markov chain Monte
Carlo for Bayesian inference, we develop a method to transform
Bayesian methods for feature allocation (and other latent structure
problems) into optimization problems with objective functions
analogous to K-means in the clustering setting. These yield
approximations to Bayesian inference that are scalable to large
Tamara Broderick is a PhD candidate in the Department of Statistics at
the University of California, Berkeley. Her research in machine
learning focuses on the design and study of Bayesian nonparametric
models, with particular emphasis on feature allocation as a
generalization of clustering that relaxes the mutual exclusivity and
exhaustivity assumptions of clustering. While at Berkeley, she has
been a National Science Foundation Graduate Student Fellow and a
Berkeley Fellowship recipient. She graduated with an AB in Mathematics
from Princeton University in 2007---with the Phi Beta Kappa Prize for
highest average GPA in her graduating class and with Highest Honors in
Mathematics. She spent the next two years on a Marshall Scholarship at
the University of Cambridge, where she received a Masters of Advanced
Study in Mathematics for completion of Part III of the Mathematical
Tripos (with Distinction) in 2008 and an MPhil by Research in Physics
in 2009. She received a Masters in Computer Science from UC Berkeley
Dr. David Lyon
Queen's University - Canada
Revelations from the National Security Agency (NSA) whistleblower Edward Snowden are making waves around the world. Mass surveillance programs track personal data from internet companies targeting everyone from ordinary citizens to heads-of-state. Many are outraged; few saw the writing on the (Facebook) wall. After commenting on (1), what exactly has been revealed, and (2) some implications, we ask how to respond, in ethical and critical ways? (3) This has been developing for decades: The rise of "risk society" and of data-driven organizations; digital dreams dominate; public-and-private blur into one. (4) Why do we tolerate it? The familiarity factor in everyday surveillance, the fear factor after 9/11 and the fun factor of social media produce compliance, not critique. (5) What's really at stake? Not just privacy and autonomy but accountability, freedom, dignity, in short, human flourishing.
Building 32 Map
Host: Jim Glass, MIT CSAIL
In recent years, "voice command" interfaces have been proposed as a means to allow drivers to engage with an expanding array of entertainment and connectivity options in the modern automobile while keeping their eyes on the road and hands on the steering wheel. A large number of studies have assessed interactions with experimental voice interfaces. Much less research is available that has examined driver behavior with production level vehicle systems. There is thus limited information on how production level voice command interfaces actually impact driver attention and how the characteristics of the interface can be optimized to better support driver attention. This talk will describe results from a series of field and laboratory studies conducted to assess driver attentional demands arising from interaction with production level voice command embedded vehicle systems and portable technologies. The field studies were conceived and implemented with the goal of developing a comprehensive assessment of the demands systems in various configurations place on drivers' attention under real-world highway driving conditions. The efforts considered drivers across a broad age spectrum. Depending on the system configuration considered, tasks assessed and measures evaluated, both positive features and possible issues associated with the attentional demands of the voice interface were identified. Overall, the results show that cognitive demands, as characterized by physiological arousal, appear lower than initial expectations. However, in certain instances, visual activity and driver orientation towards the in-vehicle display were illustrative of higher demands than might have been expected in the context of a "voice" interface. These findings suggest that there are advantages and challenges associated with voice-based interaction that need to be considered and balanced in interface design. Finally, most in-vehicle voice command interfaces should more correctly be identified as multimodal interfaces, as they typically include visual feedback and alternative response modes in addition to voice. In summary, the talk will show that drivers' interactions with these multi-modal systems can draw upon a wide array of attentional demands, often result in compensatory changes in driving behavior, and highlight the need for ongoing work to better understand the generalizability of observations. These issues are relevant in the design of in-vehicle and hand-held (smartphone and portable) technologies, and other uses of voice interactions.
Bryan Reimer, Ph.D., is a Research Engineer in the Massachusetts Institute of Technology AgeLab and the Associate Director of the New England University Transportation Center. His research seeks to develop new models and methodologies to measure and understand human behavior in dynamic environments utilizing physiological signals, visual behavior monitoring, and overall performance measures. Dr. Reimer leads a multidisciplinary team of researchers and students focused on understanding how drivers respond to the increasing complexity of the operating environment and on finding solutions to the next generation of human factors challenges associated with distracted driving, automation and other in-vehicle technologies. He directs work focused on how drivers across the lifespan are affected by in-vehicle interfaces, safety systems, portable technologies, different types and levels of cognitive load. This research also assesses the impact of medical impairments such as diabetes, cardiovascular disease, ADHD and autism. Dr. Reimer is an author on over 85 peer reviewed journal and conference papers. Dr. Reimer is a graduate of the University of Rhode Island with a Ph.D. in Industrial and Manufacturing Engineering.
This program will introduce participants to “systems thinking” as a response to the rapid changes in technology, population, and economic activity that are transforming the world, and as a way to deal with the ever increasing complexity of today's business. Systems thinking was devised to improve people's ability to manage organizations comprehensively in a volatile global environment. It offers managers a framework for understanding complex situations and the dynamics those situations produce. Senior managers can use the system dynamics method to design policies that lead their organizations to high performance. The program is intended to give participants the tools and confidence to manage organizations with full understanding and solid strategy.
The program will offer a new way of thinking about and resolving complex, persistent problems that emerge from change. Applying organization theory along with intuitive principles of feedback control, participants will learn to:
* Assess the likely impact of different policies and decisions that relate to their organization's growth, stability, and performance
* Recognize business system archetypes that can trigger persistent, long-term problems
* Use state-of-the-art management tools to identify relationships
* Intervene effectively to make fundamental changes
This program is designed for executives with decision-making responsibility who are looking for fresh ideas to resolve organizational problems. Past participants have included:
* VPs and EVPs
* Corporate planners and strategists
* Senior Project Managers
* Product Development Managers
J. Bradley Morrison
MIT Engineering Systems Division
Brad Morrison studies dynamically complex problems in organizations, organizational change, and management using the tools of system dynamics. His research centers on why organizations find it difficult to do what they want to do. Morrison focuses on implementation problems, which he has studied in several contexts, such as process improvement settings and firms adopting the practices of lean manufacturing. He tries to understand why some cases lead to successful implementation, while others lead to failures. For example, why do apparently well-intended actions often lead to outcomes that differ greatly from people?s intentions? How do the actions some managers take foster the very problems they are attempting to solve? His research is strongly rooted in organizational theory, with a methodological emphasis on interpretation through the lens of system dynamics.
Over a 20-year career with a leading management consulting firm, Morrison has assisted dozens of organizations that wrestle with change in areas such as product development and supply chain management. His consulting clients have included agencies of the United States and other governments, global consumer products firms, major retailers, and professional services firms. He has extensive experience in Asia, having worked in 11 countries on projects for clients from North America, Asia, and Europe.
Morrison teaches at MIT in the System Design and Management program, the Leaders for Manufacturing program, Executive Education programs at the MIT Sloan School of Management, and the Undergraduate Practice Opportunities Program. He is a senior scientist in the Pre-Conflict Anticipation and Shaping research team at MIT. Morrison teaches courses in business dynamics, operations management, and supply chain management in the MBA program at Brandeis University?s International Business School.
He holds a PhD in management (system dynamics and organization studies) from the MIT Sloan School of Management, an MBA in finance from the University of Chicago Graduate School of Business, and undergraduate degrees in chemistry and management science from MIT. (less)
Jay W. Forrester Professor of Management
Professor of System Dynamics and Engineering Systems
Director, MIT System Dynamics Group
John D. Sterman’s research centers on improving managerial decision making in complex systems. He has pioneered the development of “management flight simulators” of corporate and economic systems. These flight simulators are now used by corporations and universities around the world. His recent research ranges from the dynamics of organizational change and the implementation of sustainable improvement programs to experimental studies assessing the public’s understanding of global climate change. Sterman‘s research includes systems thinking and organizational learning, computer simulation of corporate strategy, and the theory of nonlinear dynamics.
He is the author of many scholarly and popular articles on the challenges and opportunities facing organizations today, including the book, Modeling for Learning Organizations, and the award-winning textbook, Business Dynamics. His articles on the innovative use of interactive simulations in management education and corporate problem solving have appeared in Fortune, the Financial Times, BusinessWeek, as well as other newspapers and journals. He has been featured on Public Television’s News Hour, National Public Radio’s Marketplace, and CBS television.
Sterman twice has been awarded the Jay W. Forrester Prize for the “Best Published Work in System Dynamics.” He also has won a 2005 IBM Faculty Award as well as the 2001 Accenture Award for the “Best Paper of the Year” published in the California Management Review (with Nelson Repenning). Five times, he has won awards for “Teaching Excellence” from the students of MIT Sloan, and was named one of MIT Sloan’s “Outstanding Faculty” by the 2001 BusinessWeek Guide to the Best Business Schools.
A great idea does not guarantee great profits. If a company's R&D dollars are going to pay off in profitable products and technologies, it needs a strategy that not only makes markets, but also beats the competition. This program will present a depth of challenges that extend from R&D to manufacturing, engineering, project management, and new ventures, and provide an innovative and powerful approach to developing technologies and products that people want to buy. The program material will also explore ways to link those technologies and products with a company's business strategy.
Drawn from MIT Sloan School's top-ranked MBA curriculum, this groundbreaking program will provide a framework for understanding how technologies and markets evolve; how they are linked; how technologies differ across markets; and how new technologies get accepted. This program will enable participants to:
* Identify profitable projects for their research dollars and find out how to capture the value of those projects
* Build technical capabilities for products that create value for their customers
* Restructure their organizations to respond to market and technical dynamics
* Implement their strategies for maximum benefit
This program is essential for senior general and technical executives involved in developing, managing, or marketing technology or products, or with managing organizations that sell their products in rapidly changing markets. The program will be most beneficial for:
* Managers in technology-intensive organizations
* Marketing and business development executives in technology organizations
* R&D managers in any organization
Titles of past participants have included:
* Executive VP
* Head of R&D; Engineering; Manufacturing & IS
* VP of Marketing & New Venture Development
* Chief Technologist
* Corporate Planner; Strategists
Building 32 Map
Host: Angelika Amon