|
|
Hossein Najafi
Professor
Discipline: Computer Science and Information Systems
Office: 212C South Hall
Phone: 715-425-3335
Email: hossein.najafi@uwrf.edu
Research Interests:
Neural Networks, Expert Systems and Fuzzy Logic and their Application to Real World Problems such as Data Modeling, Data Mining, Control Systems, Risk Management, Coding and Network Traffic Profiling are major focus of my current and past research. I have successfully applied fuzzy logic in automated tuning of control systems. A fuzzy inference system is designed to identify acceptable operating parameters of a plant with dynamic behavior. As part of this effort, we are also investigating automatic generation of the fuzzy inference system itself.
Another area of research that I am currently involved in is application of AI and data mining tools for hedging revenue risk. The volatility of commodity prices, such as crude oil prices, creates revenue risk that requires active management. In this research neural networks and other data mining tools are utilized to identify the hedging ratios that would optimize for risk and revenue.
I am also currently involved in the design and implementation of a neural network based data infusion system. Using audio perceptual coding concepts a neural network is trained to extract the auditory system’s perceptual map. This network is then used to insert perceptually masked data into an audio stream. Instrument classification in a music audio stream is another current area of research that I am currently involved in. Here data mining classification tools are used to identify a separate individual instrument used in a music audio.
The use of Artificial Neural Networks for improving performance of existing audio reconstruction filters is another area of research that I have been involved in. The basic idea is to create a reconstruction filter that would dynamically change its behavior based on the type of signals it encounters. The neural network will be trained to recognize certain important signal properties and adjust filter coefficients to match these signal properties for best reconstruction performance.
I have also been involved in application of neural networks in the area of Process Modeling, Optimization and Control. Today's manufacturing exists in a relatively data-rich, but knowledge-poor environment. An accurate predictive mode of the plant dynamics could be used to signal potential alarm conditions, reduce fluctuations in product quality, and optimize control set-points to improve yield and reduce costs. In this project, neural networks were used as function approximators that fit the process input-output noisy data with a high-dimensional surface (i.e., regression models).
I have successfully applied Neural Networks to the practical problem of Non-parametric Regression Analysis. This work has resulted in an internationally known new algorithm called Constrained Topological Mapping (CTM), that is capable of learning an unknown functional dependency from a number of noisy data points. This neural algorithm is very general and flexible, and compares favorably with traditional non-parametric regression techniques.
I also have used Neural Networks for Automating the task of Knowledge Extraction from Raw Data. Today, there exist many databases containing unprocessed raw data. However, there are few tools to extract useful information from such databases. The objective of this work was to perform automated knowledge extraction from a raw database of records.
Publications & Presentations:
- Najafi, H. L. (2010). A Neural Network Approach to Digital Data Hiding Based on the Perceptual Masking Model of the Human Vision System. International Journal of Intelligent Computing and Cybernetics, 3.
- Najafi, H. L. (2010). Modeling the Perceptual Masking Properties of the Human Auditory System Using Neural Networks. In Press, In Derick Fiedler and Rowland Krause (Ed.) Deafness, Hearing Loss and the Auditory System. Hauppauge, NY: Nova Science Publishers, Inc.
- Najafi, H. (2007). A Neural Network Approach to Audio Data Hiding Based on Perceptual Masking Model of the Human Auditory System. Journal of Applied Intelligence, 27, 269-275.
- Najafi, H. L., Rahgozar, (2006). Hedging Energy Price Risk Using Artificial Neural Networks. Accepted for publication in Journal of the Academy of Finance.
- Najafi, H. L., Rahgozar, R., & Champlin, B. (2005). An Artificial Neural Network For Hedging Crude Oil. Forthcoming in Journal of Business and Economics Research, 3 (1), 1-6.
- Najafi, H. & Champlin, B. (2005). A Fuzzy Inference System for Sub-Optimal Tuning of PID Controllers of a Simulated Autopilot. IASTED International Conference on Control and Applications.
- Najafi, H. L., Rahgozar, R., & Champlin, B. (2004). Managing Crude Oil Price Risk Using Artificial Neural Networks. IASTED International Conference on Applied Simulation and Modeling (ASM 2004).
- 2003: Najafi, H. & Champlin, B., "A Fuzzy Inference System for Automated Tuning of a Simulated Autopilot," National Conference on Undergraduate Research, held in Salt Lake City, Utah, 2003.
- Rahgozar, R. & Najafi, H. (2003). Effect of Diversification on Managing Revenue Risk. Journal of Derivatives Use, Trading, and Regulation, 9 (2), 133-149.
- Rahgozar, R. & Najafi, H. L. (2003). An Empirical Comparison of Hedging Models in Managing Oil Revenues and Risk. Journal of the Academy of Finance, 1 (101), 100-113.
Patents:
- 2004 - Triangulation Locating Listener and Artist. (# Provisional Application - No number assigned yet)
- 1996 - Dynamic Digital Filter Using Neural Networks. (# U.S. patent # 5,532,950.)
|
