If you would like to receive announcements of these
talks, please subscribe to the
cs-seminars mailing list or subscribe to the CS Seminars RSS feed ![[RSS feed]](../Common/feed-icon16x16.png)
.
Transparent Anonymization: Thwarting Adversaries Who Know the Algorithm 15 Apr 2008
Xiaokui Xiao
Chinese University of Hong Kong
Tuesday, April 15, 2008, 11:00AM
Babbio 304
Stevens Institute of Technology
Abstract
The digitization of our daily lives has led to unprecedented collections of sensitive personal data (e.g., census data, medical records) by governments and corporations. Such data is often released for research purposes, which, however, may pose a risk to individual privacy. To address this issue, numerous techniques have been proposed to anonymize the data before its publication. Somewhat surprisingly, all existing anonymization techniques assume that the adversary has no or limited knowledge of the anonymization algorithm, and fail to protect privacy when this assumption does not hold. In other words, a data publisher that adopts these techniques must take up the difficult responsibility of keeping the algorithm confidential, which severely limits the applicability of these techniques in practice.
In this talk, I will present a solution that remedies the above problem. I will start from an analytical model for evaluating disclosure risks, against an adversary who knows everything in the anonymization process, except the data to be published. This model leads to a privacy principle, transparent l-diversity, which ensures privacy protection against the adversary we consider. I will discuss three algorithms that achieve transparent l-diversity, and demonstrate their effectiveness and efficiency with experimental results.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
An Online Learning Approach to Edge Matching 28 Mar 2008
Yanghai Tsin
Siemens Corporate Research
Thursday, April 3, 11:00AM
Babbio 110
Stevens Institute of Technology
Abstract
Except in rare tightly controlled environments, many sophisticated computer vision algorithms can hardly survive the harsh visual world and its dynamic variations. Difficult to model visual scenes, imprecise sensor models, and spurious filter responses can easily contribute to failure of an otherwise mathematically elegant vision algorithm. The problem grows even harder when a real-time, highly accurate and robust system is in the specifications. Two such examples will be given at the beginning of this talk: an intelligent robotic system for parcel sorting and an augmented reality system. They lead to a common problem in vision research: robust feature matching, specifically, edge matching. I will present an appearance learning-based method for solving this problem. Starting from an initial good match given by an oracle and after each successful match afterwards, we use a randomized forest to learn a binary posteriori probability of true/false match given intensity patterns of an edge candidate. When processing a new frame, the learned model is used to predict the true matching edge. Our experiments show that the randomized forest is able to model the high dimensional, dynamic and multi-modal posterior probability timely and accurately. Using the learning-based edge matching method, we are able to track edges and estimate poses of 3D objects in many cases that are impossible using prior art.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Decentralized Access Control: The Case of Synchronous Communication and other Topics 28 Mar 2008
Constantin Serban
Rutgers University
Monday, March 31, 11:00AM
Babbio 110
Stevens Institute of Technology
Abstract
The dependability and security of modern distributed systems, such as those supporting medical institutions and commercial enterprises, require increasingly sophisticated access control mechanisms that can impose global, enterprise-wide policies. Such policies should be sensitive to the history of interaction, and should be enforced in a scalable manner.
The first part of my talk will focus on access control with respect to synchronous communication. Conventional access control is impervious to the type of communication between the components of a system, employing the same abstractions for both message-passing and synchronous communication. I will argue that the two modes of communication require different types of regulation. I will present a model that provides: (1) the regulation of both the request and the reply, separately, but in a coordinated manner; (2) regulated timeout capability provided to clients, in a manner that takes into account the concerns of their server; and (3) enforcement on both the client and server sides.
In the second part of this talk I will briefly discuss two contributions that equally apply to both synchronous and asynchronous communication: (a) Hot Policy Updates, a mechanism for propagating policy updates in a widely distributed system, without incurring inconsistencies, and (b) I will discuss a mechanism for scalable and reliable monitoring of large, open and heterogeneous distributed applications.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Interaction-Oriented Programming 25 Mar 2008
Yu David Liu
Johns Hopkins University
Thursday, March 27, 11am
Babbio 304
Stevens Institute of Technology
Johns Hopkins University
Thursday, March 27, 11am
Babbio 304
Stevens Institute of Technology
Abstract
One might have the skills to build a garage, but will the same skills suffice for building a skyscraper? Research on software construction is not new, but scale changes everything. Next-generation software systems can easily grow into billions of lines of code. Hundreds of inter-dependent platforms, sensors, and devices are commonly involved. Continuous evolution after initial deployment is often expected. Do today's programming and deployment models suffice for building these Ultra-Large-Scale (ULS) systems? In this talk, I will illustrate why today's models are lagging behind for tackling ULS systems, and how next-generation models can catch up by shifting the design focus on modeling the interactions between software building blocks.
Specifically, this talk will describe three new models I have designed -- an object model, a component model, and an application deployment model. First I will introduce an object-oriented language called Classages. At its heart, the Classages language implements an object model radically different from that of Java's and C#'s. The new model captures the complex object interactions in a well-structured fashion, and treats them as first-class citizens. Design-level concepts such as roles and associations are directly supported on the language level, so that round-trip software engineering is facilitated. The second part of the talk focuses on Assemblages, a component system that unifies static linking, dynamic linking, and distributed communication in one compact model. It is particularly suitable for programming Internet-era applications with dynamic plug-ins and distributed communications, and can potentially impact programming in emerging domains such as sensor networks. The third model I will briefly describe is Application Buildboxes, a framework to model the entire lifecycle of an application -- such as building, shipping, installation, update, and run-time evolution – during which properties such as version compatibility are enforced. All three models are formalized, with important properties (such as type soundness) rigorously proved. All three models have been implemented: the first two as prototype compilers and the third as a configuration tool.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Infusing Usability Into Security Through Natural Language 9 Mar 2008
Umut Topkara
Monday, March 10, 11:00AM
Babbio 110
Stevens Institute of Technology
Abstract
Passwords are still the first line of defense in many computer systems, hence they have a critical place among the security vulnerabilities that involve human aspects. In the first part of the talk, I will present a scheme that reconciles the apparent contradictory requirements from most password policies: That the password should be random, and that it should be memorized and never written down. The scheme is applicable to any existing text-based password authentication service and requires neither any modification to the infrastructure, nor any out of band computing device at hand (not even a calculator). Our approach consists of generating a mnemonic sentence that helps the users remember a multiplicity of truly random passwords, which are independently selected. I will also describe an improvement on this authentication system so that it provides authentication in "input constrained environments", e.g. users with motor disabilities, small electronic devices, or non-private environments.
In the second part of my talk, I will demonstrate a text watermarking method which can be used to embed a message into a text document such that it cannot be removed without destroying the value of the document. This watermarking method allows implementation of systems that manage the metadata associated with digital documents as well as rights for digital content.
I will also give a glimpse of my current and future work in other information security applications including defense against phishing.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Structure from Data 4 Mar 2008
Philippos Mordohai
University of Pennsylvania
Thursday, March 6, 11:00AM
Babbio 304
Stevens Institute of Technology
Abstract
Obtaining structure from data is a fundamental problem in computer science. In computer vision, specifically, structure inference is especially challenging due to the loss of a dimension as the 3D world is projected on images. In the first part of the talk, I will present a fundamental approach for perceptual organization that is very robust despite the absence of strong prior assumptions and global optimization. A second challenge in structure inference from data is due to the magnitude of the datasets that have to be processed for meaningful real-world applications. In the second part of the talk, I will show methods for large-scale processing of high resolution video and range data, that are scalable to entire cities.
I will begin by presenting a fundamental approach for perceptual organization. Tensor voting is a computational framework for perceptual organization of generic tokens founded on the Gestalt principles of proximity and good continuation. While these principles are widely used in computer vision, our approach is unique for a number of reasons. Arguably, the most important of these reasons is the unified representation of all structure types, such as surfaces, curves and junctions in 3D, which facilitates interactions among tokens belonging to structures of different dimensionality. The framework is applicable to a wide range of problems in computer vision and machine learning. Among them, I will briefly describe an approach to binocular stereo matching that exploits monocular cues.
In the second part of the talk, I will describe two large-scale applications of inferring structure from large data collections. The first application is a real-time, video-based 3D reconstruction system that generates accurate, detailed models from multiple video streams captured by a moving platform. Besides robustness and efficiency, the processing pipeline features several novel reconstruction techniques tailored to large-scale processing. I will also show results from an ongoing effort to segment and recognize objects from colored point clouds captured in urban environments by terrestrial and airborne range scanners. In a few months, we have obtained automatic object extraction and rough classification on datasets that exceed one billion points.
Finally, I will briefly discuss potential directions for future research.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Interface Safety in Multilingual Software 27 Feb 2008
Gang Tan
Boston University
Monday, March 3, 11:00AM
Babbio 110
Stevens Institute of Technology
Abstract
Most real-world software systems consist of software components developed in different programming languages. For example, Sun's Java Development Kit 1.6 (JDK) contains around 2 million lines of Java code as well as 800,000 lines of C/C++ code. Unfortunately, the interface code that glues multilingual software components is a constant source of software bugs and vulnerabilities. This is demonstrated by our empirical security study of JDK 1.6, which identified O(100) bugs in the interface code between Java and C components.
We propose two novel frameworks that shed light on how to achieve interface safety in multilingual software. The first system, SafeJNI, follows the principle of language-based isolation to ensure C code respects Java's invariants. The second system, ILEA, is a general framework for performing static analysis across programming-language boundaries. By automatically extracting an approximate Java specification from C code, ILEA enables existing Java analyses to cover foreign C code.
We envision the principles that drive this work are applicable to a broader context, where software components can safely interact and exchange data even if these components are in different languages, produced by different tools, or distributed over the web.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
MOSAIC: Unified Platform for Dynamic Overlay Selection and Composition 27 Feb 2008
Yun Mao
University of Pennsylvania
Thursday, February 28, 11:00AM
Babbio 304
Stevens Institute of Technology
Abstract
The Internet faces new challenges, ranging from unwanted or harmful traffic to the increasing complexity and fragility of inter-domain routing. At the same time, new applications demand evolution for new capabilities such as mobility, content-based routing, and quality-of-service (QoS) routing. Overlay networks use the existing Internet to provide connectivity for new services, and permit deployable network evolution.
Overlay networks have not, however, addressed the full set of challenges and evolutionary needs. This is due to the lack of inter-operability among different overlays. Most overlays are targeted at vertical domains (e.g., mobility, security, reliability). However, many emerging applications and application domains have needs that are difficult to address using a single overlay.
In this talk, I present MOSAIC, a unified system that provides a declarative framework for developing, deploying, combining, and composing overlay networks. It enables (1) rapid programming and deployment of new overlay networks using data-centric declarative abstractions, (2) dynamic adaptivity to select and compose overlay networks to meet changing application needs, including bridging between overlays, stacking them in layers, and dynamically changing the layers or bridges. (3) seamless support for legacy applications within the infrastructure.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Visual Recognition in the Three-dimensional World 21 Feb 2008
Silvio Savarese
University of Illinois at Urbana-Champaign
Monday, February 25, 11:00AM
Pierce 216
Stevens Institute of Technology
Abstract
The ability to interpret the semantic of objects and actions, their individual geometric attributes as well as their spatial and temporal relationships within the environment is essential for an intelligent visual system and extremely valuable in numerous applications. In visual recognition, the problem of categorizing generic objects is a highly challenging one. Single objects vary in appearances and shapes under various photometric (e.g. illumination) and geometric (e.g. scale, view point, occlusion, etc.) transformations. Largely due to the difficulty of this problem, most of the current research in object categorization has focused on modeling object classes in single (or nearly single) views. But our world is fundamentally 3D and it is crucial that we design models and algorithms that can handle such appearance and pose variability. In the first part of the talk I introduce a novel framework for learning and recognizing 3D object categories and their poses. Our approach is to capture a compact model of an object category by linking together diagnostic parts of the objects from different viewing points. The resulting model is a summarization of both the appearance and geometry information of the object class. Unlike earlier attempts for 3D object categorization, our framework requires minimal supervision and has the ability to synthesize unseen views of an object category. Our results on categorization show superior performances to state-of-the-art algorithms on the largest dataset up to date. In the second part, I present a new framework for modeling the overall geometrical and temporal organization of scenes. This is done by learning the typical distribution of spatial and temporal relationships among elements in scenes. Our model is extremely compact and can be learned in an unsupervised fashion. Experiments demonstrate that the added ability of modeling such spatial and temporal relationships is useful in several recognition tasks, such as scene/object categorization and human action classification. I conclude the talk with final remarks on the relevance of object recognition, pose estimation, and temporal and spatial reasoning as key ingredients for a coherent geometrical interpretation of images and videos.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Improving Privacy in Distributed Constraint Optimization 19 Feb 2008
Rachel Greenstadt
Harvard University
Thursday, February 21, 11:00AM
Babbio 110
Stevens Institute of Technology
Abstract
Computers increasingly act as autonomous agents entrusted with critical tasks, sensitive data, and interactions with with other humans and machines in large, multi-agent systems. These systems often require the agents to negotiate solutions to constraint optimization problems, such as resource allocation, supply-chain negotiation, and meeting scheduling. For these problems, the constraints are often extremely sensitive, representing personal or proprietary information of the agents. However, the primary algorithms for distributed constraint optimization were not developed with privacy in mind, leading to problems in meeting the needs of users.
This talk provides an analysis of the ways in which privacy has been protected and lost in distributed constraint optimization algorithms. Based on this analysis, I present a new algorithm which augments a prominent algorithm with secret sharing techniques. My approach significantly reduces privacy loss, while introducing only minimal computational overhead. My results show that my new algorithm reduces privacy loss by 29-88% on average over previous approaches.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Searchable Symmetric Encryption: Improved Definitions and Efficient Constructions 12 Feb 2008
Seny Kamara
Johns Hopkins University
Tuesday, February 19, 11:00AM
Babbio 110
Stevens Institute of Technology
Abstract
Searchable symmetric encryption (SSE) allows a party to outsource the storage of its data to another party in a private manner, while maintaining the ability to selectively search over it. This problem has been the focus of active research and several security definitions and constructions have been proposed. In this talk we will review existing security definitions, pointing out their limitations, and propose new and stronger definitions which we prove equivalent. We then give constructions that are secure under our new definitions. Interestingly, in addition to satisfying stronger security guarantees, our constructions are more efficient than all previous constructions. Further, prior work on SSE only considered the setting where only the owner of the data is capable of submitting search queries. We consider the natural extension where an arbitrary group of parties other than the owner can submit search queries. We formally define SSE in this multi-user setting, and present an efficient construction. This is joint work with Reza Curtmola, Juan Garay, and Rafail Ostrovsky.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Category Classification using Proximity Distribution Kernels and Histogram Similarity using EMD-L1 11 Feb 2008
Haibin Ling
Siemens
Friday, February 15, 11:00AM
Babbio 310
Stevens Institute of Technology
Abstract
In this talk I will introduce my work on two topics, category classification and histogram comparison. Geometric context contains very rich information for category classification tasks. We propose using proximity distribution kernels (PDK) to robustly capture co-occurrence statistics of vector-quantized local features. As a Mercer kernel, PDK is naturally combined with a support vector machine (SVM) for recognition tasks. We tested the PDK+SVM method on three public datasets for category classification. Excellent performances have been observed in all experiments in comparison to other state-of-the-art solutions.
Traditional histogram distances (e.g. L1, L2) are known to suffer from quantization and distortion problems, which can be avoided by cross-bin distances, such as the earth mover's distance (EMD). However, cross-bin distances are usually computationally too expensive for tasks involving large amount of histogram comparisons. To this end, we propose a new algorithm, EMD-L1, which reformulates the original EMD with the L1 ground distance. Using a network flow scheme, EMD-L1 runs an order faster than the original EMD solution (Rabner et al 2000). In several experiments on histogram features from both synthetic and real datasets, EMD-L1 demonstrates promising performance in both efficiency and accuracy. I will also briefly introduce an ongoing work for learning histogram similarities using SVM and boosting.
In the rest of the talk, I will briefly introduce my other work on face verification combining gradient orientation and support vector machine, robust object representation and its application in image retrieval, and a thumbnail cropping algorithm for intelligent image browsing.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Practical Dynamic Software Updating 11 Feb 2008
Iulian Neamtiu
University of Maryland, College Park
Thursday, February 14, 11am
Babbio 110
Stevens Institute of Technology
Abstract
Software updates typically require stopping and restarting an application, but many systems cannot afford to halt service, or would prefer not to. Dynamic Software Updating (DSU) addresses this difficulty by permitting programs to be updated while they run. DSU is appealing compared to other approaches for on-line upgrades because it is general, preserves state and requires no redundant hardware.
However, despite its appeal, DSU has seen limited use in mainstream software development, because it is not yet practical for mainstream programming languages. In this talk I will present an approach and a tool suite called Ginseng for dynamically updating single-threaded and multi-threaded C programs in a safe, flexible, and efficient manner.
Ginseng compiles programs specially so that they can be dynamically patched, and generates most of a dynamic patch automatically. Ginseng also performs a series of static analyses that ensure an update will not violate 1) type-safety (while guaranteeing that data is kept up-to-date), and 2) transactional version consistency (a novel update correctness property where programmers designate blocks of code as transactions whose execution must always be attributable to a single program version).
We have used Ginseng to construct and dynamically apply patches to six substantial open-source programs: three single-threaded servers (vsftpd, OpenSSH sshd, and GNU Zebra), and three multi-threaded servers (Memcached, Icecast and Space Tyrant). We dynamically patched these programs with six months to four years' worth of releases, for a total of 40 updates. Performance experiments show that while DSU support affects application performance, the impact is not significant.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
An Adaptive Systems Perspective on Network Calculus, with Applications to Autonomic Control 4 Feb 2008
Simon Dobson
Systems Research Group
UCD Dublin Ireland
Thursday, February 7 , 11:00AM
Babbio 221
Stevens Institute of Technology
Abstract
Autonomic communications systems must demonstrate behaviour that remains correct under a range of environmental conditions. In order to gain confidence that a system will behave as intended, it is advantageous to have a formal description of the expected behaviour that can be analysed and tested for compliance with different stimuli. We describe an adaptive systems perspective on a recently-proposed formal analytic model of network behaviour. We suggest how this can be used to study the adaptive behaviour of systems, and to ensure that their adaptations maintain desired properties.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Policies and Mechanisms for Secure Information Flow in the Presence of Cryptographic Operations 13 Jan 2008
Aslan Askarov
Chalmers University of Technology
Monday, January 14, 2:00PM
Babbio 304
Stevens Institute of Technology
Abstract
Cryptographic operations are essential for many security-critical systems. Reasoning about information flow in such systems is challenging because typical (noninterference-based) information-flow definitions allow no flow from secret to public data. Unfortunately, this implies that programs with encryption are ruled out because encrypted output depends on secret inputs: the plaintext and the key. However, it is desirable to allow flows arising from encryption with secret keys provided that the underlying cryptographic algorithm is strong enough. In this work we conservatively extend the noninterference definition to allow safe encryption, decryption, and key generation. To illustrate the usefulness of this approach, we propose (and implement) a type system that guarantees noninterference for a small imperative language with primitive cryptographic operations. To facilitate reasoning about released keys, we suggest an attacker-centric model of gradual release. The attacker's knowledge is modeled by the sets of possible secret inputs as functions of publicly observable events. Among the latter we distinguish special release events; the essence of gradual release is that the knowledge must remain constant between releases. Gradual release turns out to be a powerful foundation for release policies, which we demonstrate by formally connecting revelation-based and encryption-based declassification. We also show how gradual release can be enforced by security types and effects.
Based on joint work with Andrei Sabelfeld and, in part, with Daniel Hedin
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
3D Video Mosaics for Surveillance and Inspection 3 Dec 2007
Zhigang Zhu, Professor
Department of Computer Science
City College, City University of New York
Thursday, December 6, 1:00PM
McLean 104
Stevens Institute of Technology
Abstract
We address the problem of fusing images from many video cameras or a moving video camera. The captured images have obvious motion parallax, but they will be aligned and integrated into a few mosaics that preserve three-dimensional (3D) information in a large field-of-view (FOV). We have developed a pushbroom stereo mosaic representation that can re-organize the original perspective images into a set of parallel-perspective projections with different oblique viewing angles. In addition to providing a wide field of view, mosaics with various oblique views well represent occlusion regions that cannot be seen in a usual nadir view. Stereo pair(s) can be formed from a pair of mosaics with different oblique viewing angles and thus 3D viewing and 3D reconstruction can be achieved.
Supported by AFRL, NSF, ARO, DARPA, NYSIA, CUNY and industry, this approach has been applied to a number of important surveillance and inspection applications including: (1) 3D reconstruction and moving target extraction of urban scenes via an airborne camera; (2) 3D scene presentation via a camera on a ground vehicle; (3) Under-vehicle inspection via a 1D array of cameras; and (4) 3D cargo inspection via gamma-ray imaging. Algorithms for 3D reconstruction and moving target extraction for aerial video and gamma-ray images will be discussed. Video demos and experimental results will be shown.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Relay vs. User Cooperation in Multiaccess Networks 23 Nov 2007
Lalitha Sankar
Princeton University
Monday, November 26, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
Cooperation in communication networks results when terminals use their energy and bandwidth resources to mutually enhance their transmissions. Cooperation can be induced in many ways and each approach entails a different tradeoff of power, bandwidth, complexity, and costs to achieve spatial diversity gains characteristic of antenna arrays. In this talk, we present a specific cooperative network - a multiaccess relay channel (MARC) - where cooperation is induced via a dedicated relay node in a network where multiple independent users communicate with one destination.
We first extend the classical relaying strategies of decode-and-forward (DF), compress-and-forward (CF), and amplify-and-forward (AF) to the MARC. Next we compare this approach of inducing cooperation in a multiaccess channel using a relay (relay cooperation) to the approach of allowing the users to cooperate with each other (user cooperation). Using the total transmit and processing power consumed as a cost metric, we compare the DF and AF outage probabilities for the two networks. We model processing costs as a function of transmission rates and use geometry-inclusive outage analyses as well as area-averaged numerical results to show that cooperation is most desirable in the regime where processing power is significantly smaller than the transmit power. We also show that the maximum spatial diversity gains resulting from user cooperation may not always be achievable in SNR regimes of practical interest without trading off delay and complexity. Finally, our results also reveal that relay cooperation is on average more energy efficient than user cooperation.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
A Model Curriculum for Graduate Software Engineering Education 23 Nov 2007
Arthur Pyster
Distinguished Research Professor
School of Systems & Enterprises
Stevens Institute of Technology
Monday, November 19, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
Over 50 universities in the United States and many others globally offer a Masters Degree in Software Engineering. However, the most current software engineering model graduate curriculum was developed by the Software Engineering Institute at Carnegie Mellon over 15 years ago. Given how differently today's software is used and developed, a fresh look at graduate programs is needed. A broad coalition of professionals from academia, industry, and government is creating a new model curriculum. The curriculum team is completing an initial study of the existing software engineering graduate programs that shows broad diversity in goals, content and requirements for admission and graduation. The model curriculum is being strongly influenced by the model curriculum for undergraduate software engineering education, SE2004, that was published by the IEEE and ACM and by the IEEE's Software Engineering Body of Knowledge. Most importantly, it also considers industry desires concerning the skills and competencies they expect to see in a graduate of a master's program and tries to integrate essential elements of systems engineering as well. It will provide a graduate-level core curriculum based on a common body of knowledge and be flexible enough for individual organizations to create the program that best responds to their goals, individual strengths, and target student population.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Ensemble of Classifiers Approaches for Data Fusion, Incremental Learning, Early Diagnosis of Alzheimer's Disease and Becoming a Millionaire! 23 Nov 2007
Robi Polikar, Ph.D.
Electrical and Computer Engineering, Rowan University, Glassboro, NJ
Friday, November 16, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
In matters of great importance that has financial, medical, social or other implications, we often seek a second opinion before making a decision, sometimes a third, and sometimes many more. In doing so, we somehow weigh the individual opinions, and combine them through some thought process to reach a final decision in hope of making that decision the most informed one. The process of consulting "several experts" before making a final decision what may be second nature to us has recently been rediscovered by computational intelligence community for automated decision making applications, and it has emerged as a popular and heavily researched area. Also known under various other creative names, such as multiple classifier systems, committee of classifiers or mixture of experts, ensemble systems have shown to produce favorable results compared to those of single expert systems for a broad range of applications, and under a variety of scenarios. While the design, implementation and application of such systems are the main thrusts of ensemble system research, we will focus on one family of algorithm in this talk: Learn++. Learn++ is a versatile approach that is capable of incremental learning of supplementary information provided by additional data (even if such data introduce new classes), as well as data fusion of complementary information provided by independent data sources. Following an overview of various ensemble-based approaches, Learn++ and its derivatives will be explained in detail. The performance of these family of algorithms will be presented on several applications, including early diagnosis of Alzheimer's disease from event related potentials of the electroencephalogram. Oh, as for becoming a millionaire - well, for that you will have to just come and listen the talk.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Stevens / Columbia / IBM Research Security and Privacy Day 10 Nov 2007
Stevens / Columbia / IBM Research
Security and Privacy Day
Friday, November 16, 2007
Babbio 104
Stevens Institute of Technology
Hoboken, New Jersey, USA
Web site:
http://www.cs.stevens.edu/~jcordasc/security_and_privacy_day
Registration:
http://www.cs.stevens.edu/~jcordasc/security_and_privacy_day/registration.html
[Registration is free, but requested.]
Program:
* 08:30 - 09:30 Registration and Breakfast
* 09:30 - 09:35 Opening Remarks
* 09:35 - 10:45 Keynote Talk: Andrew W. Appel (Princeton University)
* 10:45 - 11:00 Break
* 11:00 - 11:30 Sven Dietrich (Stevens Institute of Technology)
* 11:30 - 12:00 Juan Garay (Bell Labs Alcatel-Lucent)
* 12:00 - 02:00 Lunch Break (on your own)
* 02:00 - 02:30 Salvatore J. Stolfo (Columbia University)
* 02:30 - 03:00 Break
* 03:00 - 03:30 Larry Koved, Ted Habeck (IBM Research)
* 03:30 - 04:00 Antonio Nicolosi (Stevens Institute of Technology)
* 04:00 Concluding Remarks
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Parallel Algorithms for Bayesian Indoor Positioning Systems 10 Nov 2007
Konstantin Kleisouris
Computer Science Department
Rutgers University
Monday, November 12, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
In this work we present two parallel algorithms and their Unified Parallel C implementations for Bayesian indoor positioning systems. Our approaches are founded on Markov Chain Monte Carlo simulations, which explore the probability distributions of the unknown position variables using statistical sampling. We evaluated two basic partitioning strategies: inter-chain partitioning which distributes entire Markov chains to different processors, and intra-chain which distributes a single chain across processors. Evaluations on a 16-node symmetric multiprocessor, a 4-node cluster comprising of quad processors, and a 16 single-processor-node cluster, suggest that for short sampling chains intra-chain parallelism scales well on the first two platforms, with speedups of up to 12. On the other hand, inter-chain parallelism gives speedups of 12 only for very long Markov chains, sometimes of up to 60,000 samples, on all three platforms. We used the LogGP model to analyze our algorithms and predict their performance. We found the model a useful guide in our algorithm design, as well as useful for predicting performance. When running the inter-chain algorithm, the model predictions are within 5% of the actual execution times, while for intra-chain they are 7%-25% less than the measured times due to load imbalance not captured in the model.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
A Deductive Framework for Security Policy Analysis 10 Nov 2007
C. R. Ramakrishnan
Computer Science Department
Stony Brook University
Friday, November 9, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
Rule languages have been used to specify security and management policies, such as access control and authorization policies, and network management policies. While these rule languages aim to simplify the specification and management of complex policies, large rule sets often contain subtle interactions, making them difficult to understand and reason about. Deductive spreadsheets (DSS) offer a new way of manipulating rules using a familiar spreadsheet-like interface. We will explore the use of DSS for security policy analysis, in particular for analyzing information flow properties of Role-Based Access Control (RBAC) rules, and for vulnerability analysis. Incremental evaluation is a central component of DSS: as in a traditional spreadsheet, when a cell value changes, all dependent cells are incrementally updated in a DSS. This talk will also describe the incremental rule evaluation techniques that form the basis for DSS.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Cognitive Radios for License-exempt Use 10 Nov 2007
Kiran Challapali
Philips Research North America
Monday, November 5, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
Although most of the spectrum is allocated much of it is unused. The enormous growth in the wireless industry has come from using a small part of the wireless spectrum, nominally less than 10% under 3 GHz. There is growing evidence of scarcity and overcrowding in these bands reflected for example, by price paid for cellular spectrum. However, measurements have shown other parts of the spectrum - although allocated - are virtually unused, and known widely as spectrum white spaces. These white spaces vary from place to place and time to time. Cognitive Radio (CR) technologies enable harnessing these spectrum white spaces, permitted by new spectrum regulation. Specifically, spectrum regulation to allow the unlicensed use of television bands is well underway in the US.
The breakthrough provided by Cognitive Radios is significant because it allows the development of new and innovative types of devices and services for businesses and consumers, without disrupting television and other authorized services. Key benefits include: (a) access to abundant new spectrum, (b) better propagation therefore reliable, low outage communication and low power operation, and (c) peaceful coexistence with other wireless networks. Cognitive Radio topic has featured in MIT Technology Review's top ten most promising emerging technologies. Accordingly, during the last few years, there has been a lot of effort by the FCC, industry and standard groups to make CRs happen. There has also been significant research effort in academia for better understanding of the technical requirements and potential solutions.
In this presentation, we will discuss the status of spectrum regulation in allowing unlicensed use of television bands, and technical challenges and solutions to realize the potential of Cognitive Radios.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Design and Analysis of Low-rate Codes 10 Nov 2007
Guosen Yue
NEC Research Laboratories
Monday, October 29, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
We consider the design and analysis of low-rate codes, generalized low-density parity-check (GLDPC) codes and the variations, repeat-zigzag-Hadamard (RZH) codes, in AWGN channels. The code ensembles of GLDPC and RZH codes can be optimized using Extrinsic information transfer (EXIT) charts. The simulation results show that a rate-0.003 LDPC-Hadamard code built from designed code ensemble with large block length performs only 0.15 dB away from the ultimate Shannon limit (-1.592 dB). Those low-rate codes are the promising candidates for the channel coding in the applications of sensor network, ultra-wideband communications and code-spread CDMA systems.
Blinklist | Del.icio.us | Digg | Furl | Ma.gnolia | Newsvine | Spurl | Technorati
Securing Untrustworthy Software Using Information Flow Control 12 Oct 2007
Nickolai Zeldovich
Stanford University
Monday, October 22, 2:00PM
Babbio 310
Stevens Institute of Technology
Abstract
In this talk, I will present HiStar, a new operating system designed to enforce the security of user data in untrusted or malicious applications. For example, numerous web sites have had massive data compromises due to poorly-written application code. HiStar can ensure that even malicious application code in a web server cannot disclose sensitive user data.
The key idea is to specify application security in terms of information flow, or what can happen to the data, and enforce it at a narrow kerne