Ph.D. Dissertation Proposal

MLVFS: A Scalable Storage System For Managing Big Scientific Data

Navid Golpayegani

3:00-5:00pm Tuesday 24 February 2015, ITE 346

Managing peta or exabytes of data with hundreds of millions to billions of files is a necessary first step towards an effective big data computing and collaboration environment for distributed systems. Current file system designs have focused on providing better and faster data distribution. Managing the directory structure for data discovery becomes an essential element of the scalability problems for big data systems. Recent designs are addressing the challenge of exponential growth of files. Still largely unexplored is the research for dealing with the organizational aspect of managing big data systems with hundreds of millions of files. Most file systems organize data into static directory structures making data discovery, when dealing with large data sets, hard and slow.

This thesis will propose a unique Multiview Lightweight Virtual File System (MLVFS) design to primarily deal with the data organizational management problem in big data file systems. MLVFS is capable of the dynamic generation of directory structures to create multiple views of the same data set. With multiple views, the storage system is capable of organizing available data sets by differing criteria such as location or date without the need to replicate data or use symbolic links. In ad- dition, MLVFS addresses scalability issues associated with the growth of the stored files by removing the internal metadata system and replacing it with generally avail- able external metadata information (i.e. data base servers, project compute servers, remote repositories, etc.). This thesis, moreover, proposes to add, plug in capabilities not normally found in file systems that make this system highly flexible, in terms of specifying sources of meta data information, dynamic file format streaming and other file handling features.

The performance of MLVFS will be tested in both simulated environments as well as real world environments. MLVFS will be installed on the BlueWave cluster at UMBC for simulated load testing to measure the performance for various loads. Simultaneously, stable version of MLVFS will run in real world production environ- ments such as those of the NASA MODIS instrument processing system (MODAPS). The MODAPS system will be used to show examples of real world use cases for MLVFS. Additionally, there will be other systems explored for the real world use of MLVFS, such as at NIST for research into Biomedical Image Stitching.

Committee: Drs. Milton Halem (Chair, Advisor), Yelena Yesha, Charles Nicholas, John Dorband, Daniel Duffy

Understanding Social Spammers: A Data Mining Perspective
Xia “Ben” Hu

Computer Science and Engineering
Arizona State University

12:00-1:00 Tuesday, 24 February 2015

With the growing popularity of social media, social spamming has become rampant on all platforms. Many (fake) accounts, known as social spammers, are employed to overwhelm legitimate users with unwanted information. Social spammers are unique due to their coordinated efforts to launch attacks such as distributing ads to generate sales, disseminating pornography and viruses, executing phishing attacks, or simply sabotaging a system’s reputation. In this talk, I will introduce a novel and systematic analysis of social spammers from a data mining perspective to tackle the challenges raised by social media data for spammer detection. Specifically, I will formally define the problem of social spammer detection and discuss the unique properties of social media data that make this problem challenging. By analyzing the two most important types of information, network and content information, I will introduce a unified framework by collectively using heterogeneous information in social media. To tackle the labeling bottleneck in social media, I will show how we can take advantage of the existing information about spam in email, SMS, and on the web for spammer detection in microblogging. I will also present a solution for efficient online processing to handle fast-evolving social spammers.

Xia Hu is a Ph.D. candidate in Computer Science and Engineering at Arizona State University, supervised by Professor Huan Liu. His research interests include data mining, machine learning, social network analysis, etc. As a result of his research work, he has published nearly 40 papers in several major academic venues, including WWW, SIGIR, KDD, WSDM, IJCAI, AAAI, CIKM, SDM, etc. One of his papers was selected for the Best Paper Shortlist in WSDM’13. He is the recipient of IEEE “Atluri Award” Scholarship, 2014 ASU’s President’s Award for Innovation, and Faculty Emeriti Fellowship. He has served on program committees for several major conferences such as WWW, IJCAI, SDM and ICWSM, and reviewed for multiple journals, including IEEE TKDE, ACM TOIS and Neurocomputing. His research attracts wide range of external government and industry sponsors, including NSF, ONR, AFOSR, Yahoo!, and Microsoft.

– more information and directions: http://bit.ly/UMBCtalks –

Topic Modeling with Structured Priors for Text-Driven Science

Michael Paul, JHU

12:00pm – 1:00pm, Monday, 2 March 2015, ITE 325

Many scientific disciplines are being revolutionized by the explosion of public data on the web and social media, particularly in health and social sciences. For instance, by analyzing social media messages, we can instantly measure public opinion, understand population behaviors, and monitor events such as disease outbreaks and natural disasters. Taking advantage of these data sources requires tools that can make sense of massive amounts of unstructured and unlabeled text. Topic models, statistical models that describe low-dimensional representations of data, can uncover interesting latent structure in large text datasets and are popular tools for automatically identifying prominent themes in text. However, to be useful in scientific analyses, topic models must learn interpretable patterns that accurately correspond to real-world concepts of interest.

In this talk, I will introduce Sprite, a family of topic models that can encode additional structures such as hierarchies, factorizations, and correlations, and can incorporate supervision and domain knowledge. Sprite extends standard topic models by formulating the Bayesian priors over parameters as functions of underlying components, which can be constrained in various ways to induce different structures. This creates a unifying representation that generalizes several existing topic models, while creating a powerful framework for building new models. I will describe a few specific instantiations of Sprite and show how these models can be used in various scientific applications, including extracting self-reported information about drugs from web forums, analyzing healthcare quality in online reviews, and summarizing public opinion in social media on issues such as gun control.

Michael Paul is a PhD candidate in Computer Science at Johns Hopkins University. He earned an M.S.E. in CS from Johns Hopkins University in 2012 and a B.S. in CS from the University of Illinois at Urbana-Champaign in 2009. He has received PhD fellowships from Microsoft Research, the National Science Foundation, and the Johns Hopkins University Whiting School of Engineering. His research focuses on exploratory machine learning and natural language processing for the web and social media, with applications to computational epidemiology and public health informatics.

– more information and directions: http://bit.ly/UMBCtalks –

Last week, health insurance giant Anthem revealed that the personal information of as many as 80 million customers was stolen by hackers. This news came just days before President Obama announced the creation of a new agency to analyze and counter cyber threats. In this hour, we look at Obama’s cybersecurity agenda, and the cyber-security challenges that face users in the coming year.

Joining UMBC’s Cybersecurity Graduate Program Director for this morning’s hour-long discussion were Joseph Marks, Cybersecurity Reporter, Politico Pro (his article on the new cybersecurity center) and Dakota S. Rudesill, Assistant Professor of Law, Moritz College of Law The Ohio State University

(Source)

MP3 Podcast: Download (Duration: 49:57 — 68.6MB)

ACM Tech Talk

Studying Internet Latency via TCP Queries to DNS

Dr. Yannis Labrou
Principal Data Architect, Verisign

1:30-2:30pm Friday, 27 February 2015, ITE 456, UMBC

Every day Verisign processes upwards of 100 billion authoritative DNS requests for .COM and .NET from all corners of the earth. The vast majority of these requests are via the UDP protocol. Because UDP is connectionless, it is impossible to passively estimate the latency of the UDP-based requests. A very small percentage of these requests though, are over TCP, thus providing the means to estimate the latency of specific requests and paths for a subset of the hosts that interact with Verisign’s network infrastructure.

In this work, we combine this relatively small number of datapoints from TCP (on the order of a few hundred million per day) with the much larger dataset of all DNS requests. Our focus is the process of data analysis of real world, imperfect data at very large scale with the goals of understanding network latency at an unprecedented magnitude, identifying large volume, high latency clients and improving their latency. We discuss the techniques we used for data selection and analysis and we present the results of a variety of analyses, such as deriving regional and country patterns, estimations for query latency for different countries and network locations, and techniques for identifying high latency clients.

It is important to note that latency results we will report are based on passive measurements from, essentially, the entire Internet. For this experiment we do not have control over the client side — where they are, which software, their configuration, their network congestion. This is significantly different from latency studied in any active measurement infrastructure such as Planet Lab, RIPE Atlas, Thousand Eyes, Catchpoint, etc.

Dr. Yannis Labrou is Principal Data Architect at Verisign Labs where he leads efforts to create value from the wealth of data that Verisign’s operations generate every day. He brings to Verisign 20 years of experience in conceiving, creating and bringing to fruition innovations; combining thinking big with laboring through the pains of materializing ideas. He has done so in an academic environment, at a startup company, while conducting government and DoD/DARPA sponsored research and for a global Fortune 200 company.

Before joining Verisign, Dr. Labrou was a Senior Researcher at Fujitsu Laboratories of America, Director of Technology and member of the executive staff of PowerMarket, an enterprise application software start-up company and a Research Assistant Professor at UMBC. He received his Ph.D. in Computer Science from UMBC, where his research focused on software agents, and a Diploma in Physics from the University of Athens, Greece. He has authored more than 40 peer-reviewed publications, with almost 4000 citations and he has been awarded 14 patents from the USPTO. His current research focus is data through the entire lifecycle from generation to monetization.

– more information and directions: http://bit.ly/UMBCtalks –

Baltimore’s Walters Art Museum will host ArtBytes 3, its third annual hackathon, this weekend from 6pm Friday to 3pm Sunday. Technology and creative communities will work together at the museum to build programs and applications that enhance the museum experience for visitors. $5,000 in cash prizes will be awarded.

It’s a great opportunity for students to try their hand at helping to conceive, debug and prototype an ideas for the common good. For more information and to register, see the ArtBytes 3 EventBright page.

The Fall 2014 CMPE Town Hall Meeting has been rescheduled for Friday, February 13, at 12 noon in ITE 104.  Pizza will be available in the atrium outside ITE 104 at the conclusion of the meeting.

The meeting will have a limited agenda, due to time constraints. A second Town Hall meeting will be happening later in the semester, to address issues such as Capstone and departmental Undergraduate Research opportunities.

Agenda:

• Welcome and Introductions  — Dr. LaBerge
• Changes to the CMPE Program — Dr. LaBerge
• The BS/MS Program — Dr. LaBerge
• Open forum questions and answers

If you responded to the Google Docs survey in 2015 (as opposed to the form for Fall 2014), please respond to the link below.

To ensure that there is enough pizza, please fill out the following Google Docs form if you plan on attending:
https://docs.google.com/a/umbc.edu/forms/d/1ia6FY86BWvnO_eomrzaseMbCSn_h9HKBoFB3c2iYr2g/viewform?usp=send_form.

MS Thesis Defense

Real-time Realistic Rendering of Sunrise and Sunset on the Ocean

Yuping Zhang

10:00am Monday, 9 February 2015, ENG 005

The aim of this thesis is to study advanced real-time realistic rendering techniques for outdoor natural scenes. Much research has been done for rendering realistic outdoor scenes, such as sky, ocean, terrain, etc. But sunrise and sunset are rarely discussed. Interesting phenomena like the sun mirages during sunset and sunrise could create splendid visual effects.

This thesis presents a method to render sun mirages. It uses precomputed atmospheric refraction profiles for different atmosphere condition and performes ray tracing to render the sun deformation. Combined with other methods, it renders sunrise and sunset on the ocean in real time. It simulates realistic sun mirages, sky color, ocean waves and lighting, which can be useful for realistic scenes in movies, video games or for scientific simulators.

Program Committee: Drs. Marc Olano (Chair), Penny Rheingans, Jian Chen

The Computational Photonics Laboratory has one or two post-doctoral/research professor positions available.  Please click for more information: http://www.csee.umbc.edu/?p=18210.

The Computational Photonics Laboratory has one or two post-doctoral/research professor positions available.  Please click for more information: http://www.csee.umbc.edu/?p=18210.