Kurt Luther, Ph.D.
Carnegie Mellon University
Tuesday, January 21, 2014
Social technologies have given rise to "crowd creativity," connecting people from around the world to share ideas, pool resources, leverage diverse skill sets, and produce compelling, sometimes world-changing creative artifacts. Crowd creativity offers the potential to make the future of work more rewarding, productive, and equitable, but tapping this potential requires us to understand its strengths and limitations, and how to design for them. In this talk, I present some of my recent efforts to push the boundaries of crowd creativity, focusing on two major challenges: helping crowds coordinate, and helping them perform tasks requiring expertise. I will describe several studies of "collabs," a genre of collaborative animation projects with especially complex coordination issues, and Pipeline, a creative collaboration tool I designed, deployed, and evaluated with the goal of helping leaders coordinate their projects. I will also discuss the challenges of using novice crowd workers in domains requiring specialized expertise, focusing on CrowdCrit, a system I developed to provide visual designers with fast, scalable, high-quality critiques. Finally, I will present some preliminary work supporting crowd creativity for knowledge discovery. Throughout the talk, I will suggest ways to design for more complex, creative, and successful social computing experiences.
Kurt Luther is a postdoctoral fellow in the HCI Institute at Carnegie Mellon University. His research explores the intersection of crowds, computing, and creativity across a range of application areas, including computer animation, visual design, and knowledge discovery. Many of his projects, including Pipeline and ProveIt, involve building and evaluating creativity support tools released as open-source software. Kurt completed his Ph.D. in Human-Centered Computing at Georgia Tech, where he received the GVU Center's highest honor, the Foley Scholarship. He has worked in YouTube's User Experience group and the Social Computing groups at Microsoft Research and IBM Research. Kurt's research has been recognized by the Best Paper Award at CSCW 2013 and featured in TIME, The Atlantic, and Harvard Business Review, among others.
Hosted by Leysia Palen.
Jessica Hullman, Ph.D.
University of California, Berkeley
Thursday, January 16, 2014
A shift in the availability of usable tools and public data has prompted mass manufacturing of information visualizations to communicate data insights to broad audiences. Studying professional practice suggests that expert visualization designers and analysts negotiate complex design trade-offs in creating visualizations: decisions between competing design goals that may call for different representations, data selections, or levels of complexity. I will describe several tools and approaches for modeling common trade-offs in order to support more efficient creation and use of data visualizations in online contexts. This includes systems that automate design decisions to produce customized, annotated interactive visualizations to add context to news articles. Successful visualizations also strike a balance between presenting data as subject to uncertainty to support accurate interpretation and depicting data in ways that can be understood by users without statistical background. I will describe a technique for visualizing uncertainty as hypothetical data samples, and present experimental results around the potential benefits of this technique for conveying the reliability of patterns in data. Taken together, these results provide principles and tools that scale effective visual data communication to larger numbers of contexts and users.
Jessica Hullman is a postdoctoral fellow in the Computer Science department at the University of California Berkeley. Her research addresses challenges related to the diverse use of visualizations for communication in online contexts. Her work provides design tools and systems for topics in data storytelling and presentation, including the use of visualizations for supporting understanding of the news and for communicating complex analyses to non-analysts. Her research also addresses visualization reception in social and other online contexts, including challenges related to uncertainty visualization and social and other contextual biases factors affecting group analysis settings. Jessica has received multiple awards and honors for her research, including an Honorable Mention for Best Paper at IEEE InfoVis 2011, and the Gary M. Olson Outstanding Ph.D. student award at the University of Michigan. She has conducted visualization research in the V.I.B.E. group at Microsoft Research and the C.U.E. group at I.B.M. Research.
Hosted by Leysia Palen
Qin (Christine) Lv
Assistant Professor, University of Colorado Boulder
Thursday, July 18, 2013
Big data is becoming the norm in a wide range of application domains, be it science, engineering, education, commerce, national security, or people's daily information needs. Typically characterized by the 4Vs (volume, velocity, variety, and veracity), big data is continuously generated yet human's capabilities are very limited in terms of digesting such massive and complex data for information exploration and knowledge discovery. This calls for a paradigm shift that goes beyond traditional storage and retrieval of raw data and aims to make sense of data at scale. Through the integration of efficient system design and effective data analysis techniques, along with domain specific knowledge, our goal is to discover and exploit useful patterns in big data with high quality and high efficiency.
This talk gives an overview of my research on big data analytics, which bridges the areas of ubiquitous computing, data mining, mobile systems, social networks, and data management. Two specific themes of my research will be presented. For data‐oriented scientific discovery, working with domain experts, my research centers on data management and data analysis in environmental sensing of air pollutants, time series analysis of cryospheric data, and energy storage system design for transportation electrification. For data oriented mobile social computing, I have investigated distributed mobile data management, social data analysis and recommendation, and video chat in online and mobile settings.
Qin Lv is an Assistant Professor in the Department of Computer Science, University of Colorado Boulder. She received her Ph.D. in Computer Science from Princeton University in 2006. Before joining CU, she also spent one year in Princeton University as a postdoc, and one year in the Computer Science Department of Stony Brook University as an Assistant Professor.
Lv's research integrates efficient system design and effective data analysis for the management and exploration of big data. Her research spans the areas of ubiquitous computing, data mining, mobile systems, social networks, and data management. Her research is interdisciplinary in nature and interacts closely with a variety of research domains including environmental science, geosciences, renewable and sustainable energy, materials science, as well as the information needs in people's daily lives. Lv has over 40 peer-¬‐reviewed publications in many top venues including UbiComp, MobiSys, Pervasive, WWW, KDD, SIGMETRICS, etc. She has one Best Paper Award nomination, and her work on personalized driving behavior monitoring and analysis for emerging hybrid vehicles won the Computational Sustainability Award bestowed jointly by the Computing Community Consortium (CCC) and Pervasive 2012.
Hosted by Richard Han.
Jordan Boyd-Graber, University of Maryland
Thursday, April 11, 2013
ABSTRACT: A common information need is to understand large, unstructured datasets: millions of e-mails during e-discovery, a decade worth of science correspondence, or a day’s tweets. In the last decade, topic models have become a common tool for navigating such datasets. This talk investigates the foundational research that allows successful tools for these data exploration tasks: how to know when you have an effective model of the dataset; how to correct bad models; how to scale to large datasets; and how to detect framing and spin using these techniques. After introducing topic models, I argue why traditional measures of topic model quality---borrowed from machine learning---are inconsistent with how topic models are actually used. In response, I describe interactive topic modeling, a technique that enables users to impart their insights and preferences to models in a principled, interactive way. I will then address computational and statistical limits to existing approaches and how streaming topic models, with an "infinite vocabulary", can be applied to real-world online datasets. Finally, I’ll discuss ongoing collaborations with political scientists to use these techniques to detect spin and framing in political and online interactions.
BIO: Jordan Boyd-Graber is an assistant professor in the University of Maryland's iSchool and the Institute for Advanced Computer Studies. He is a 2010 graduate of Princeton University, with a PhD thesis on "Linguistic Extensions of Topic Models", under David Blei.
Tom Erez, University of Washington
Tuesday, March 19, 2013
ABSTRACT: Science-fiction robots can perform any task humans do and more. In reality, however, today's articulated robots are disappointingly limited in their motor skills. Current planning and control algorithms cannot provide the robot with the capacity for intelligent motor behavior - instead, control engineers must manually specify the motions of every task. This approach results in jerky motions (popularly stereotyped as "moving like a robot") that cannot cope with unexpected changes.
I study control methods that automate the job of the controls engineer. I give the robot only a cost function that encodes the task in high-level terms: move forward, remain upright, bring an object, etc. The robot uses a model of itself and its surroundings to optimize its behavior, finding a solution that minimizes the future cost. This optimization-based approach can be applied to different problems, and in every case the robot alone decides how to solve the task. Re-optimizing in real time allows the robot to deal with unexpected deviations from the plan, generating robust and creative behavior that adapts to modeling errors and dynamic environments.
In this talk, I will present the theoretic and algorithmic aspects needed to control articulated robots using model-based optimization. I will discuss how machine learning can be used to create better controllers and my work on trajectory optimization. Finally, I will describe my vision for the future of model-based optimization in robotics and control. A preview of some of the work discussed in this talk can be seen here: dl.dropbox.com/u/57029/MedleyJan13.mp4 [a lower-quality version is also available on YouTube: youtube.com/watch?v=t4JdSklL8w0]
BIO: Tom Erez is a Postdoctoral Research Associate in the Computer Science and Engineering department at the University of Washington. He currently leads the controls group in the the UW team for the DARPA Robotics Challenge, building a system for autonomous control of a humanoid robot in a disaster response scenario. His research is concerned with intelligent motor control of robotic and bio-mechanical systems; in particular, he is interested in the interface between machine learning and control theory. His past publications include work on gait optimization, physics simulation, planning under uncertainty, modeling hand-eye coordination, minimax control, and real-time optimization for control. He was awarded the Sackler Scholarship for his research on bio-mechanical models, and his work has been funded by NSF, DARPA, and Willow Garage. Tom received his PhD in Computer Science from Washington University in St. Louis, and his BSc in Mathematics from the Hebrew University in Jerusalem. He also spent three years as a researcher at the University of Amsterdam and the Institute for Scientific Interchange in Turin, Italy, where he studied the application of Statistical Physics models to domains of natural and social sciences.
Hosted by Dirk Grunwald