RESEARCH AREAS

Data Mining and Knowledge Discovery
Data mining (DM) is the process of extracting relevant knowledge from large and complex data sets. A broader process, Knowledge Discovery (KD) (also refereed in a limited scope as knowledge discovery from databases (KDD)) involves preprocessing (data preparation), DM and post processing (knowledge refinement) processes. DM algorithms work on pre-processed data and knowledge refinement processes validate and refine discovered knowledge. My interests in DM and KD revolve around both basic and applied research. In particular, I am interested in developing fundamentally new techniques that build on soft computing, pattern matching, statistical, and biological systems. For the applied aspects, I am interested in applications of DM and KD to Internet security. Of special interest are anomaly detection in World Wide Web and computer networks to identify intruders and malicious activity and, detection and control of malicious executables.

Fault Mitigation in Software Systems
The search for fundamental principles of fault tolerance in human-engineered complex dynamic systems is very new. We are interested in modeling complex dynamic systems as hybrid interacting automata whose continuously varying dynamics capture the physical process at the lowest level of abstraction. Discrete event models at the higher levels capture the cognitive response of the system to observed emerging physical phenomena. Our broader aim is to formulate analytical models of the higher-level dynamics of component interactions triggered by all types of individual failures to (i) predict emerging pathological system behavior from time-series observations of events and their dynamic interactions, and (ii) formulate adaptive mechanisms to circumvent or mitigate the effects of pathological behavior.

In particular, my interests are mitigation of faults in complex software systems. One area that we are currently exploring is to model interactions of software applications with the Operating System as a deterministic finite state automaton (DFSA) (i.e., a regular language) and apply the Supervisory Control Theory for development of a recognizer of this language to control and mitigate faults in software execution. Specifically, the discrete-event supervisor restricts the legal language of the model in an attempt to mitigate the normal detrimental consequences of faults or undesirable events.

Web and Internet Security
Web caching, Web site reorganization, Web personalization, Web optimization, Trust and related protocols. My interests are design and implementation, security aspects of the World Wide Web at the system level. I am also interested in design of algorithms, Internet protocols, Internet programming, study of Internet attack methods, and building software based Internet security mechanisms above the IP layer. I am also interested in authentication and trust protocols and methods, Internet security policies, mobile code control of malicious software, and intrusion detection.

Soft Computing
Soft Computing (SC) consists of Fuzzy Logic, Neural Computing, Evolutionary Computation, Machine Learning, and Probabilistic Reasoning, which include belief networks, chaos theory and parts of learning theory. SC tolerates imprecision, uncertainty, partial truth, and approximation to achieve tractability, robustness and low solution cost. SC has been influenced by a lot of earlier work but I attribute the most influence on SC (in its present form) to Zadeh's 1965 paper on fuzzy sets; the 1973 paper on the analysis of complex systems and decision processes; and the 1979 report (1981 paper) on possibility theory and soft data analysis. The inclusion of neural computing and genetic computing in soft computing came at a later point. SC is a foundational component for the emerging field of conceptual intelligence.

I am interested in doing basic and applied work in soft computing with applications to the Internet, World Wide Web, and computer networks. Earlier in my research, I have applied two organizing principles of functional architecture of mammalian primary visual cortex: (1) Competitive learning and (2) Hebb type learning to image restoration and segmentation. I am interested in extending this work to apply to steganography (hiding messages in Web images, etc.)

World Wide Web and Computer Networks
This area includes the ability of the network to monitor itself, find alternate paths for traffic through the network and modify these paths dynamically based on the state of the network. The specific topics of interest in this area are: (1) Load estimation based on the history of network traffic at node level. Considerable success has been achieved in using recurrent neural nets to predict chaotic time series like those based on Glass-Mackey equation. The advantage of these techniques is that hardware implementation can result in real time load estimation. (2) Adaptive route selection. In case of failure of particular channel or path, fast rerouting of traffic is very important. Number of possible routes in a densely connected network (or a wide area network with a number of LANs acting as nodes or even mobile units acting as nodes) can increase combinatorialy. The selection of paths will also depend upon the importance of the message, the existing traffic and the topology of the network. I want to apply here some of the emerging techniques using Neural Nets (Hopfield net and Kohonen's feature maps) to find optimal path in a dynamic environment; I am also interested in exploring the Genetic algorithms and Fuzzy logic, these techniques may have potential.