Computational Intelligence: models, theory and applications

The advanced level of today's research in Computational Intelligence makes the practical applications of such methods feasible. Research at SeaLab has been covering this area for more than ten years. The results obtained can be arranged within the following macro categories:

Research in this area mainly concerns innovative algorithms for intelligent data processing. Classification and unsupervised data representation are the basic areas of interest. The basic goal always is to develop algorithms that can prove effective in real-world applications; therefore a major accent is posed on the characterization of the expected performance, thereby involving generalization and validation issues. The research in this area lead to the formulation of some novel models of Connectionist Computing:

• Circular Back-Propagation (CBP) Networks;
• Conditional Class-Entropy (CCE) Networks for classification;
• Plastic Neural Gas (PGAS) training algorithm for Vector Quantization;
• K-Winner Machines (KWMs).

The following table summarizes the various applications that have been experienced for the above-mentioned models, typically including high-dimensional intelligent signal processing:

Methods and Theoretical Models	Applications
Connectionist Models for intelligent signal processing Circular Back-Propagation Networks	Image-quality prediction Industrial process control and pollutant-emission forecasting Clinical diagnosis Object visual tracking
Connectionist Models for data clustering and mining Vector Quantization	Network Security License-plate localization Optical Character Recognition Novelty detection Text Mining for (business) intelligence
Support Vector Machines for classification	Image-quality prediction Data mining, novelty detection Security Risk-assessment scenarios
Advanced Computational Paradigms Quantum Computing Optimization	Advanced methods for optimal engineering design

The research on applied Quantum Computing aims at exploiting Quantum Parallelism in solving highly complex design problems in engineering.
The ability of Quantum Computers to handle an exponential amount of information using a linear amount of resources allows one to tackle optimization task that otherwise would only involve empirical, sub-optimal approaches. The research has applied Quantum Optimization methods for the optimal training of several paradigms from Computational Intelligence, namely, quantum optimization for Support Vector Machines and Quantum training of neural Vector Quantizers.