SSS Abstracts  Fall 2009

Presented by Terrill L. Frantz, ISRI-COS

An often overbearing logistical problem that social network researchers and analysts face is the challenge of carrying out the basic data processing steps necessary to transform real-world source data into a form that is formatted specifically to the requirements of particular social network analysis software. In particular, this talk focuses on resolving the variety of real-world data forms that must be transformed into the format necessities of the software suite developed at the Center for Computational Analysis of Social and Organizational Systems (CASOS) at Carnegie Mellon University. This data processing or programming problem, while rather straightforward, can be daunting to non-I.T. users of these ¡ªand any-- software tools, and at best case, laboriously costly for the programming savvy. This talk outlines and describes a software architecture and user paradigm to address this ubiquitous problem.

BBM: Bayesian browsing model from petabyte-scale data

Tuesday, October 13^th, 2009 from 12-1 pm in GHC 4303.

Presented by Fan Guo, CSD

Given a petabyte-scale web search click log, can we build scalable probabilistic models to effectively interpret user clicks and estimate user-perceived relevance? In this talk, I will review previous studies on click position-bias, discuss key hypotheses which act as building blocks for click models, and introduce the Bayesian browsing model and its inference algorithm with desired properties of being closed-form, single-pass and parallelizable. Experimental results to test model effectiveness and efficiency will also be presented.

(Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.)

Identifying Social Network Subgroups using ORA

Tuesday, October 20^th, 2009 from 12-1 pm in GHC 4303.

Presented by Terrill L. Frantz, ISRI-COS

This presentation is a primer on using ORA software for identifying social network subgroups according to traditional and newly-formulated techniques, such as Johnson Hierarchical Clustering, CONCOR, Girvan-Newman, FOG, and others. This material will be delivered in a non-technical manner with minimal coverage on the underpinnings of the techniques, and will focus on the usability of ORA to perform these subgrouping tasks. For general information about ORA, see www.casos.cs.cmu.edu/projects/ora/.

Integrating Multiple-Subject Multiple-Study fMRI Datasets Using Canonical Correlation Analysis

Friday, October 23^rd, 2009 from 12-1 pm in GHC 4303.

Presented by Indrayana Rustandi, CSD

Predictive approaches for fMRI data analysis that integrate fMRI data from multiple subjects and multiple studies are desirable because they can potentially leverage more data to make better predictions, and in addition, inform us about similarities and differences that exist across different human subjects and different studies. However, in order to be successful, these approaches must deal with variations that exist in different brains both anatomically and functionally. In this talk, I present an approach to integrate multiple fMRI datasets in the context of predictive fMRI data analysis. The approach utilizes canonical correlation analysis (CCA) to find common dimensions among the different datasets, implicitly taking into account the anatomic and functional variations present in the data. We apply the approach to the task of predicting brain activations for unseen concrete-noun words using multiple-subject datasets from two related fMRI studies. The proposed approach yields better prediction accuracies than those of an approach where each subject's data is analyzed separately.

(Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.)

Simulating Organizational Networks using Construct

Friday, October 30^th, 2009 from 12-1 pm in GHC 4303.

Presented by Terrill L. Frantz, ISRI-COS

This presentation is a primer on using Construct software for simulating social behavior in an organization. Construct is a multi-agent model of group and organizational behavior. Human beings are inherently variable and complex. Construct models groups and organizations as complex systems and captures the variability in human and organizational factors. The non-linearity of the model generates complex temporal behavior due to dynamic relationships among agents. These dynamic relationships are grounded in structuration theory which is the notion of construction and reconstruction of the social system through human interaction based on rules and resources. The changes in the social system are defined and analyzed through the lens of social network analysis. For general information about Construct, see www.casos.cs.cmu.edu/projects/construct/.

No Authorization Without Representation: Conveying the authority applications require users to grant as a condition of installation

Friday, November 6^th, 2009 from 12-1 pm in GHC 4303.

Presented by Jennifer Tam, CSD

Computer operating systems, and now websites that serve as application platforms, are increasingly adopting stricter application security models; they restrict the resources applications can access to those authorized by the user. Users authorize access to these resources either when the application is installed or when the application rst requires access to an unauthorized resource. While the security of users systems and data increasingly rests on their ability to make these authorization decisions, there is little research to guide those designing these authorization experiences.

We performed a laboratory study to evaluate different designs for representing the actions and resources to be authorized as a condition of installing an application. We used a within-participants design to observe thirty-three Facebook users ability to absorb and search information in seventeen different representations, all of which were presented in the context of a fictional Facebook application. Four of these representations conveyed only a set of resources to be authorized, such as contacts or friends. The other thirteen representations paired resources with different actions that could be performed on them, such as seeing information, changing information, or adding new information.

We find that participants overwhelmingly prefer representations in which resources are presented visually, using icons or pictures. We also find strong evidence that users are able to search presentations containing icons more quickly than those that do not. Finally, we found that participants performed better when authorization information was organized by actions, and followed by the various resources on which the actions would be authorized, than when information was grouped by the resources.

Joint work with Stuart Schechter (MSR) and Robert Reeder (Microsoft)

(Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.)

Usability of Robots that Ask for Help

Tuesday, November 17^th, 2009 from 12-1 pm in GHC 4303.

Presented by Stephanie Rosenthal, CSD

Robots are often given tasks to assist humans in their environments. A task-driven robot communicates with humans to clarify its state and actions. The humans are often experts or designated supervisors of the robot and, while they provide very accurate responses to the robot, it is expensive to maintain such a pairing of robots and humans. Instead, we propose a more symbiotic human-robot relationship, in which humans are benefited by the robot's autonomous actions and the robot can ask any humans in the environment for help to complete its tasks. While the symbiotic relationship can be mutually beneficial, it can also be a detriment - reducing the usability as well as the accuracy of the human help.

My work focuses on improving the usability of such task-driven robots that employ symbiotic human-robot interaction, while still aiming to improve the accuracy of non-supervisor responses. Specifically, through varying the timing of the robot's questions and the information it provides humans about its state while asking for help. We model both the information needs of the non-supervisor to answer a robot's questions accurately as well as the perceived utility of a symbiotic robot in terms of the value of the tasks and cost of answering questions. We present user studies to evaluate the usability of a symbiotic robot that 1) provides helpful information to humans with the questions it asks and 2) uses the expected utility to determine when to ask and which actions to take. We conclude with current work implementing the results on a functional robot.

Joint work with Manuela Veloso and Anind K. Dey.

(Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.)

Ditto: Eavesdropping for the Common Good in Multi-hop Wireless Networks

Friday, December 4^th, 2009 from 12-1 pm in GHC 4303.

Presented by Amar Phanishayee, CSD

This talk presents the design, implementation, and evaluation of Ditto, a system that opportunistically caches overheard data to improve subsequent transfer throughput in wireless mesh networks. While mesh networks have been proposed as a way to provide cheap, easily deployable Internet access, they must maintain high transfer throughput to be able to compete with other last-mile technologies. Unfortunately, doing so is difficult because multi-hop wireless transmissions interfere with each other, reducing the available capacity on the network. This problem is particularly severe in common gateway-based scenarios in which nearly all transmissions go through one or a few gateways from the mesh network to the Internet.

Ditto exploits on-path as well as opportunistic caching based on overhearing to improve the throughput of data transfers and to reduce load on the gateways. It uses content-based naming to provide application independent caching at the granularity of small chunks, a feature that is key to being able to cache partially overheard data transfers. Our evaluation of Ditto shows that it can achieve significant performance gains for cached data, increasing throughput by up to 7x over simpler on-path caching schemes, and by up to an order of magnitude over no caching.

This was work done with Fahad Dogar, Himabindu Pucha, Olatunji Ruwase, and David Andersen. Ditto was published in the proceedings of ACM MOBICOM 2008.

(Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.)

Improving Prediction of fMRI Data By Integrating Multiple Subjects and Studies

Friday, December 11^th, 2009 from 12-1 pm in GHC 4303.

Presented by Indrayana Rustandi, CSD

Predictive approaches for fMRI data analysis that integrate fMRI data from multiple subjects and multiple studies are desirable because they can potentially leverage more data to make better predictions, and in addition, inform us about similarities and differences that exist across different human subjects and different studies. However, in order to be successful, these approaches must deal with variations that exist in different brains both anatomically and functionally. In this talk, I present an approach to integrate multiple fMRI datasets in the context of predictive fMRI data analysis. The approach utilizes canonical correlation analysis (CCA) to find common dimensions among the different datasets, implicitly taking into account the anatomic and functional variabilities present in the data. We apply the approach to the task of predicting brain activations for unseen concrete-noun words using multiple-subject datasets from two related fMRI studies. The proposed approach yields better prediction accuracies than those of an approach where each subject's data is analyzed separately.

Joint work with Tom Mitchell and Marcel Just.

(Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.)

Web contact: sss+www@cs