Adaptive Neural Signal Processing Systems

Starting from nonlinear adaptive systems based on neural networks, the objective is to study methods for:
* model evaluation and interpretation
* adaptive learning in non-stationary environments
* optimization of model structures
* design of experimental conditions including database design.
Model evaluation (including generalization ability) and interpretation are fundamental issues when designing signal processing systems for practical applications, and several problems regarding definition and reliable estimation are still to be solved.
The fact that most practical problems involves adaptation to changing
environmental conditions calls for investigation of methods for model
design, including optimization of model structure. In particular, recurrent
neural networks and heterogeneous network ensembles will be studied. Finally,
the project covers methods for experimental design, especially active learning and combined supervised/unsupervised learning schemes.
The theoretical research is carried out in close synergy with application
projects covering:
* Analysis and interpretation of brain scan data
* Medical decision support systems,
* Humanitarian mine detection
* Monitoring and inspection systems.