WIE Undergraduates: CS Scholars Projects
2008 Summer Research Projects
Please note that "Daily Activities" are provided as estimates of time commitment and do not include additional program activities.
- Quantification of Computer Security
Faculty Mentor: Dr. Michel Cukier, Mechanical Engineering - Importance of Prosody in Speaker Identification and Verification
Faculty Mentor: Dr. Carol Espy-Wilson, Electrical & Computer Engineering - Sensor Development for Increased Reliability in Monitoring Devices
Faculty Mentor: Dr. Kristine Rosfjord, Electrical & Computer Engineering - Detection of Simulated Human Users with Auditory CAPTCHAs
Faculty Mentor: Dr. Jonathan Simon, Electrical & Computer Engineering - Detecting Attacks with Anomaly-based Intrusion Detectors
Faculty Mentor: Dr. Kymie Tan
Quantification of Computer Security
Project Description
Most research in computer security focuses on developing methods and tools for preventing, detecting, or tolerating attacks or intrusions. Little research is conducted to quantify the security of a computer system. The Experimental Information Assurance Lab (EIAL) at the University of Maryland specializes in conducting experiments for quantifying various security aspects of a computer system. Recent research focused on: a) how to separate attacks among malicious traffic, b) how to analyze malicious activity over time, c) how to assess the threat of attacks due to IRC channels, d) how to understand behavior and diagnosing and finding solutions to issues encountered in organizational computer security systems using a systemic approach, namely system archetypes, e) How to build a profile of attacker behavior following a remote compromise, and f) how to analyze a large set of incident data using well-known tools developed by the software reliability community.
The Summer 2008 project focuses on modeling various parts of a large public university including attacks, networks, and users. More precisely, the focus will be on a large set of incident data (i.e., over 12,000 records collected over 6 years) and attack data collected by Intrusion Prevention Systems (IPSs) (i.e., about 4000 alerts a day since September 2006). The analysis and modeling activity will address among others: a) the correlation between incidents and IPS alerts, b) the evolution of the IPS data as a function of the attack type, and c) the evolution of the IPS data as a function of the location. This research will be an iterative process between model refinement and data collection and analysis to feed the models.
CS Scholar Responsibilities and Daily Activities
All CS Scholars will be expected to perform background reading (10%) on analysis methods and malicious traffic data collection. All Scholars will be trained on the test-bed deployed for collecting and analyzing malicious traffic (20%). Each scholar will analyze data and develop a model (65%). The Scholars will meet at least once a week to compare experiences and identify specific activities for ongoing study (5%). All Scholars will compile their results for a presentation and report publication.
Importance of Prosody in Speaker Identification and Verification
Project Description
O
One area of research in the Speech Communication Lab focuses on the development of robust speaker identification and speaker verification systems. We have primarily focused on the extraction of low-level acoustic information related to a speaker’s voice quality (i.e., breathy, creaky or modal) or the type of vocal tract configuration a speaker uses when producing certain complicated sounds. In addition to this type of speaker specific information, we want to explore the use of prosody in distinguishing between speakers. We know from impersonators that it is important to get the “rhythm” of the imitated speaker correct in order to be successful. In this project, we will study the relationship between various acoustic parameters and the prosodic information conveyed in the speech signal.
CS Scholar Responsibilities and Daily Activities
This project will require the student to do some background reading on speaker recognition (10%), studying present recognition codes (15%), perform some statistical analysis of the relationship between prosodic events and values of acoustic parameters (35%), implementation of algorithm to take capture prosodic features (20%) and testing of algorithm (20%). All Scholars will compile their results for a presentation and report publication.
Sensor Development for Increased Reliability in Monitoring Devices
Project Description
Sensors integrated into monitoring and security devices require high levels of reliability and repeatability. Surface enhanced Raman scattering (SERs) substrates that are capable of obtaining chemical specific information thus require signal enhancement that is reproducible across the substrate and repeatable from substrate to substrate. These characteristics are not prevalent in the SERS substrates available today.
The summer 2008 project will offer the student the opportunity to develop the experimental skills necessary for substrate characterization. Specifically, the student will learn the optical techniques necessary for testing substrates as well as how to analyze this experimental data. This data acquisition and analysis will help steer the nanofabrication process development of these in-house fabricated substrates.
CS Scholar Responsibilities and Daily Activities
All CS Scholars will be expected to perform background reading (10%) on SERs and optical testing techniques. All Scholars will be trained in the optical laboratory (20%). Each scholar will characterize substrates and analyze substrate data (65%). The Scholars will meet at least once a week to compare experiences and identify specific activities for ongoing study (5%). All Scholars will compile their results for a presentation and report publication.
Detection of Simulated Human Users with Auditory CAPTCHAs
Project Description
Automatic methods to distinguish human users from simulated users are critical to withstand security attacks which degrade the quality of service of large multiuser systems, whether due to abuse or resource expenditure. CAPTCHAs (Completely Automated Public Turing test to tell Computers and Humans Apart) can be deployed to protect systems vulnerable to e-mail spam, such as common web-mail services. (see http://en.wikipedia.org/wiki/CAPTCHA for more details). Currently most CAPTCHAs are visual: users are instructed to retype a piece of highly distorted text—very robust against computer attacks but not too difficult for typical human users. Those CAPTCHAs cannot be deployed for visually-impaired users, however, nor is it guaranteed that computer algorithms will not soon be up to the task of reading them. As an alternative, auditory CAPTCHAs are also now deployed, where users are given the option of listening to a series of distorted spoken digits.
The Summer 2008 project aims to provide the student with a strong grasp of the state-of-the-art in automated speech recognition, its potential, and its limitations. Specifically, the students will analyze state-of-the-art auditory CAPTCHAs for both human and computer intelligibility, to isolate weaknesses and identify new directions for improvement. Additionally, we hope to identify methods of degrading speech (or other easily identifiable sounds) in ways that are especially difficult for automatic speech recognition methods but not so for human listeners.
CS Scholar Responsibilities and Daily Activities
All CS Scholars will be expected to perform background reading (10%) on visual and auditory CAPTCHAs and speech recognition. All Scholars will implement and deploy novel methods for the creation of auditory CAPTCHAs (20%). All Scholars will be trained to design evaluation strategies, analyze and compare the weaknesses and strengths of different methods of creating auditory CAPTCHAs, from both the computer and human points of view (65%). The Scholars will meet at least once a week to compare experiences and identify specific activities for ongoing study (5%). All Scholars will compile their results for presentation and report publication.
Detecting Attacks with Anomaly-based Intrusion Detectors
Project Description
Novel attacks and insider threats are among the more pressing concerns in the area of cyber-warfare. The detection of such undesirable events typically falls in the purview of anomaly-based intrusion detection systems. Unfortunately, these detectors are accompanied by problems that make them less effective than desired, for example, prohibitively high false alarm rates, a high degree of uncertainty regarding the true extent of a detector’s performance space and a poor understanding of how and why detector performance is affected by various parameter values.
CS Scholar Responsibilities and Daily Activities
All CS Scholars will be expected to perform background reading (10%) on anomaly-based intrusion detection and experimental methodology. All Scholars will implement and deploy detection algorithms (20%). All Scholars will be trained to design evaluation strategies, analyze and compare detector capabilities (65%). The Scholars will meet at least once a week to compare experiences and identify specific activities for ongoing study (5%). All Scholars will compile their results for presentation and report publication.
Return to CS Scholars | Projects Overview
