Program Faculty

Research: Prof Blaze's research focuses on the architecture and design of secure systems based on cryptographic techniques, analysis of secure systems against practical attack models, and on finding new cryptographic primitives and techniques. This work has led directly to several new cryptographic concepts, including: "Remotely-Keyed Encryption," which allows the use of inexpensive, low-bandwidth secure hardware to protect high-bandwidth communication and stored data, "Atomic Proxy Cryptography," which allows re-encryption by untrusted third parties, and "Master-Key Encryption," which provides a systematic way to design (and study) ciphers with built-in "back doors."

Prof Blaze is especially interested in the use of encryption to protect insecure systems such as the Internet. He was a designer of swIPe, a predecessor of the now standard IPSEC protocol for protecting Internet traffic. Another project, CFS, investigated and demonstrated the feasibility of including encryption as file system service.

Roch Guerin
Electrical & Systems Engineering

Quality-of-Service (QoS), with a focus on scalable and flexible mechanisms that require minimimum configuration and interactions between the network and users. However, in general and in spite, or rather because of many years investigating QoS topics, I have reached the conclusion that the large majority of QoS solutions turn out to be more expensive than the resources they are trying to manage. Hence, the bulk of my current activities are directed at what I would characterize as:
Robust Networking: What I mean by that are a set of techniques that allow you to design networks and network mechanisms that can ensure efficient operation across a broad range of operational charactetistics. For example, in the traffic engineering area, this means identifying routing schemes that are tolerant of link and node failures as well as changes in traffic patterns, in the sense that they result in good overall network performance even in the presence of such perturbations. In the more traditional QoS area, this means devising mechanisms that support service differentiation across a broad range of traffic characteristics, i.e., are not heavily dependent on the proper configuration of policers. I am also interested in extending the concept of robust networking to the wireless setting, where the combination of greater limitations on resources and the more dynamic nature of users and of the network infrastructure itself, creates a new set of problems.

Dwight L. Jaggard
Electrical & Systems Engineering

Electromagnetic Chirality: Examination of the use of electromagnetic chirality for applications to scattering, absorption, radiation, and devices.
Fractal Electrodynamics: Investigation of the integration of electromagnetic waves with multiscale fractal structures.
Scattering from Knots: Utilization of the polarization of waves to remotely investigate and characterize the handedness and knottedness of conducting loops.
Imaging and Inverse Scattering: Development of approximate and exact imaging algorithms for image and texture classification.

Statistical Processing: Development of nonlinear signal processing techniques based on radial basis function networks and other neural network structures for applications in communications, interference rejection, image restoration, recognition, and prediction. Application of order statistics concepts in multivariate signal and image processing and biomedical signal analysis and processing.
Wireless Communications: Techniques to combat impairments in wireless channels through use of special diversity, coding, modulation, and detection strategies.

Research interests are in the general area of communications networks with emphasis on resource allocation, routing and scheduling, optimization and control of stochastic systems, distributed systems and algorithms, quality of service and pricing issues.

Research is currently directed towards the implementation of 1 Gbps WANs. Approach taken is to view the network as a support fabric for distributed shared memory (DSM). Memory is then used by processes to communicate as if they were located on the same machine. The advantage is that the software overhead of communication is greatly reduced from the traditional scenario of a protocol stack implemented as if the network fabric was two cans connected by a piece of string. DSM requires protection semantics similar to that of local memory, which provides read/write protection on a process-by-process basis. The inclusion of networks, however, invalidates the typical assumption of complete physical control of the machine, since the "memory bus" may have been extended into other administrative domains. The use of cryptography and authentication protocols allows a robust (i.e., including access control) DSM to be implemented, albeit with some performance penalty. By integrating cryptography and caching, much of the overhead can be optimized away.

The next step is to find higher-performance authentication protocols. Competitive Parallel Processing uses the idea of exploiting multiple possibilities for a solution to obtain speedup with parallel processors. The idea relies on randomness in execution time, and a number of theorems relating the degree of randomness to speedup have recently been proven and experimentally validated. Follow-on work extends the analysis to other measures besides execution time, and uses the analysis of speedup to improve stopping criteria in searches.

Neural Networks: Computational complexity; randomized learning algorithms.
Computational Learning Theory and Information Theory: Model complexity, optimal stopping, and regularization; learning to detect fraud; threshold functions.
Pattern Recognition: Finite-sample performance of algorithms; utilizing side information; the information content of an example.

Research: Prof Zdancewic studies programming languages and computer security. Most recently, his work has focused on language-based enforcement of information-flow policies. He is also interested in secure concurrent and distributed computing, functional programming languages, type theory, and linear logic.