Presidential Young Investigator Award: Application of Neural Networks to Controller Design Using Feedback Linearization

The P.I. will study the analysis and design of large- scale stochastic systems, specifically binary neural networks and adaptive distributed detection systems. He will work on a theoretical framework for these systems, and on demonstration of the application of the theory in assessing existing designs and in devising synthesis techniques. The expected applications are in Machine Intelligence and in Sensor Fusion. The objective is to provide designers and users of neural networks and distributed-detection systems with a set of reliable polynomial-time algorithms, for calculating in advance (at least a bound on) the expected performance. In particular, such assessments are important when the designer using suboptimal or 'intelligent' (i.e. heuristic) design algorithms. Often, a close-form optimal architecture is not available, or the implementation such design is prohibitive from the computational or the hardware- complexity viewpoints. A related objective is to synthesize and assess on- line learning algorithms, since the common assumptions on availability of environmental statistics and on stationary of the statistics are often not realistic.