Select a calendar:
Filter April Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for April
-
PhD Defense - Rui Miao
Wed, Apr 11, 2018 @ 12:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Rui Miao
Committee: Minlan Yu (Chair), Ramesh Govindan, Konstantinos Psounis
Title: Scaling-out Traffic Management in the Cloud
Abstract:
Managing cloud traffic is challenging due to its large and constantly growing traffic in scale and traffic anomalies. Network infrastructure and traffic management need to scale their capacity to such traffic growth and anomalies, or the application performance will suffer. Existing traffic management functions have so far focused on proprietary hardware appliances and software servers. However, with limited capacity and/or fixed functionality, those solutions incur a high cost, low performance, and high management complexity.
In this thesis, we argue that we should scale-out traffic management functions for the full throughput of datacenter networks. The key idea of this thesis is to leverage the hardware switches with line-rate packet processing and the emerging programmability to directly build advanced functionaries. We have scaled-out three major traffic management functions: load balancing, attack mitigation, and congestion control. Our evaluation shows a high performance and cost-efficiency from our solutions.
Location: Charles Lee Powell Hall (PHE) - 631
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Amulya Yavdav
Thu, Apr 12, 2018 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Amulya Yadav
Committee: Milind Tambe (Chair), Kristina Lerman, Aram Galstyan, Eric Rice, Dana Goldman
Title: Artificial Intelligence for Low Resource Communities: Influence Maximization in an Uncertain World
Time: April 12 (Thursday) 1:00-3:00 PM
Location: KAP 209
Abstract:
The potential of Artificial Intelligence (AI) to tackle challenging problems that afflict society is enormous, particularly in the areas of healthcare, conservation and public safety and security. Many problems in these domains involve harnessing social networks of under-served communities to enable positive change, e.g., using social networks of homeless youth to raise awareness about Human Immunodeficiency Virus (HIV) and other STDs. Unfortunately, most of these real-world problems are characterized by uncertainties about social network structure and influence models, and previous research in AI fails to sufficiently address these uncertainties, as they make several unrealistic simplifying assumptions for these domains.
This thesis addresses these shortcomings by advancing the state-of-the-art to a new generation of algorithms for interventions in social networks. In particular, this thesis describes the design and development of new influence maximization algorithms which can handle various uncertainties that commonly exist in real-world social networks (e.g., uncertainty in social network structure, evolving network state, and availability of nodes to get influenced). These algorithms utilize techniques from sequential planning problems and social network theory to develop new kinds of AI algorithms. Further, this thesis also demonstrates the real-world impact of these algorithms by describing their deployment in three pilot studies to spread awareness about HIV among actual homeless youth in Los Angeles. This represents one of the first-ever deployments of computer science based influence maximization algorithms in this domain. Our results show that our AI algorithms improved upon the state-of-the-art by 160% in the real-world. We discuss research and implementation challenges faced in deploying these algorithms, and lessons that can be gleaned for future deployment of such algorithms. The positive results from these deployments illustrate the enormous potential of AI in addressing societally relevant problems.
Location: Kaprielian Hall (KAP) - 209
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Sonal Mahajan
Thu, Apr 19, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Sonal Mahajan
Time: April 19, 2018 (Thursday) 9.30am - 12pm
Location: VHE 217
Committee: William G. J. Halfond (chair), Nenad Medvidovic, Sandeep Gupta, Chao Wang, Jyotirmoy Vinay Deshmukh
Title: Automated Repair of Presentation Failures in Web Applications Using Search-based Techniques
Abstract:
The appearance of a web application's User Interface (UI) plays an important part in its success. Issues degrading the UI can negatively affect the usability of a website and impact an end user's perception of the website and the quality of the services that it delivers. Such UI related issues, called presentation failures, occur frequently in modern web applications. Despite their importance, there exist no automated techniques for repairing presentation failures. Instead repair is typically a manual process were developers must painstakingly analyze the UI of a website, identify the faulty UI elements (i.e., HTML elements and CSS properties), and carry out repairs. This is labor intensive and requires significant expertise of the developers.
My dissertation addresses these challenges and limitations by automating the process of repairing presentation failures in web applications. My key insight underlying this research is that search-based techniques can be used to find repairs for the observed presentation failures by intelligently and efficiently exploring large solution spaces defined by the HTML elements and CSS properties. Based on this insight, I designed a novel general-purpose search-based framework for the automated repair of presentation failures in web applications. To evaluate the effectiveness of my framework, I designed and developed instantiations for repairing different types of presentation failures in web applications. The first instantiation focuses on the repair of Cross Browser Issues (XBIs), i.e., inconsistencies in the appearance of a website when rendered in different web browsers. The second instantiation addresses the Mobile Friendly Problems (MFPs) in websites, i.e., improves the readability and usability of a website when accessed from a mobile device. The third instantiation repairs problems related to internationalization in web application UIs. Lastly, the fourth instantiation addresses issues arising from mockup-driven development and regression debugging. In the empirical evaluations, all of the four instantiations were highly effective in repairing presentation failures, while in the conducted user studies, participants overwhelmingly preferred the visual appeal of the repaired versions of the websites compared to their original (faulty) versions. Overall, these are positive results and indicate that the framework can help developers repair presentation failures in web applications, while maintaining their aesthetic quality.
Location: Vivian Hall of Engineering (VHE) - 217
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Siddharth Jain
Mon, Apr 23, 2018 @ 11:30 AM - 01:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Siddharth Jain
Committee: Ron Artstein (Chair), Paul Rosenbloom, Morteza Dehghani, Kallirroi Georgila, Elsi Kaiser
Title: Identifying Social Roles in Online Contentious Discussions
Time: Mon, April 23, 2018 @ 11:30 AM - 1:30 PM.
Room: SOS B45.
Abstract:
The main goal of this dissertation is to identify social roles of participants in online contentious discussions, define these roles in terms of behavioral characteristics
participants show in such discussions, and develop methods to identify these participant roles automatically. As social life becomes increasingly embedded in online systems, the concept of social role becomes increasingly valuable as a tool for
simplifying patterns of action, recognizing distinct participant types, and cultivating and managing communities. In contentious discussions, the roles may exert a major influence on the course and/or outcome of the discussions. The existing work on social roles mostly focuses on either empirical studies or network based analysis. Whereas this dissertation presents a model of social roles by analyzing the content of the participants' contribution. In the first portion of this dissertation, I present the corpus of participant roles in online discussions from Wikipedia: Articles for Deletion and 4forums.com discussion forums.
A rich set of annotations of behavioral characteristics such as stubbornness, sensibleness, influence, and ignored-ness, which I believe all contribute in the identification of roles played by participants, is created to analyze the contribution of the participants. Using these behavioral characteristics, Participant roles such as leader,
follower, rebel, voice in wilderness, idiot etc. are defined which reflect these behavioral
characteristics. In the second part of this dissertation I present the methods used to identify these participant roles in online discussions automatically using the
contribution of the participants in the discussion. First, I develop two models to identify leaders in online discussions that quantify the basic leadership qualities of participants. Then, I present the system for analyzing the argumentation structure of comments in discussions. This analysis is divided in three parts: claim detection, claim delimitation, and claim-link detection. Then, the dissertation presents the social roles model to identify the participant roles in discussions. I create classification models and neural network structures for each behavioral characteristic using a set of features based on participants' contribution to the discussion to determine the behavior values for participants. Using these behavioral characteristic values the roles of participants
are determined based on the rules determined from the annotation scheme. I show that for both, the classification models and neural networks, the rule based methods perform
better than the model that identifies the participant roles directly. This signifies that the framework of breaking down the problem of identifying social roles to determining
values of specific behavioral characteristics make it more intuitive in terms of what we expect from participants who assume these roles. Although the neural network methods
perform worse than their traditional classification method counterparts, when provided with additional training data, neural network structures improve at a much higher rate.
In the last part of the dissertation, I use the social roles model as a tool to analyze participants' behavior in large corpus. Social roles are automatically tagged in
Wikipedia corpus containing -26000 discussions. This allows determining participants' roles over time in order to identify whether they assume different roles in different discussions, and what factors may affect an individual's role in such discussions. I investigate three factors: topic of the discussion, amount of contention in the discussion, and other participants in the discussion. The results show that participants behave similarly in most situations. However, the social roles model is able to identify instances where participants' behavior patterns are different than their own typical behavior. In doing so, the model provides useful context regarding the reason behind these behavioral patterns by identifying specific behaviors affected by the situation.Location: Social Sciences Building (SOS) - B45
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Stephanie Kemna
Mon, Apr 30, 2018 @ 02:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Stephanie Kemna
Committee: Gaurav Sukhatme (chair), Nora Ayanian, David Caron
Title: Multi-Robot Strategies for Adaptive Sampling with Autonomous Underwater Vehicles
Time & place: Monday April 30th, 2pm, RTH406
Abstract:
Biologists and oceanographers are sampling lakes and oceans worldwide, to obtain data on the natural phenomena they are interested in. For example, measuring algae abundance to understand and explain potentially harmful algal blooms. Typical methods of sampling are (a) taking physical water samples and sensor measurements from boats, (b) deploying sensor packages off of buoys, docks or other static man-made structures, and more recently (c) running pre-programmed missions with aquatic robots. The use of robot teams could significantly improve cost- and time-efficiency of lake and ocean sampling, allowing persistent and efficient mapping of the water column in finer resolution. Additionally, these systems may be able to intelligently gather data without needing a lot of prior information. We envision a scenario where one or two groups of biologists or oceanographers come together for monitoring a lake, bringing their autonomous vehicles with biological sensors.
Our focus is on improving sampling efficiency, and environmental modeling performance, through the addition of (decentralized) coordination approaches for multi-robot sampling systems. In this presentation, I will discuss adaptive informative sampling techniques for single- and multi-robot deployments. Adaptive informative sampling means that the robots adapt their trajectory online, based on sampled data, while incorporating information-theoretic metrics to seek out the most informative sampling locations. Through simulation studies we have shown the benefits that can be obtained from employing adaptive informative sampling approaches. We include field results to show the feasibility of running adaptive informative sampling on board an autonomous underwater vehicle (AUV).
For the multi-robot case, we show the benefits that can be obtained from adding data sharing between vehicles, and we explore the trade-off of surface based (Wi-Fi) communications versus underwater (acoustic) communications. In terms of coordinating multiple vehicles, I will first discuss an explicit coordination approach, based on dynamic estimation of Voronoi partitions, which shows potential for improving modeling performance in the early stages of model creation. I then discuss a method we developed for how to best start adaptive sampling runs when no prior data is available. Finally, I will discuss the use of implicit coordination through asynchronous surfacing with a surface-based data hub. We showed that performance across surfacing strategies was similar, though some turned out to be less consistent in performance, and some methods showed potential for greatly reducing the number of surfacing events needed.
Overall, I have developed several methods for adaptive informative sampling with AUVs, focusing on multi-robot coordination and field constraints. The results of my studies show the benefits and potential of incorporating data sharing and coordination strategies into adaptive sampling routines for multi-robot systems.
Location: Ronald Tutor Hall of Engineering (RTH) - 406
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
