IJCNN 2005 - Verification, Validation and Certification of Neuro-Adaptive Controllers in Safety-Related Areas
(in conjunction with IJCNN 2005 International Joint Conference on Neural Networks Montreal, Canada)

August 5th 2005

Over the recent years, artificial Neural Networks (NNs) have found their way into various safety-related and safety-critical areas, like transportation, avionics, environmental monitoring and control, and medical applications. Quite often, these applications (using NN techniques ranging from classification to monitoring and control) proved to be highly successful, leading from a pure research prototype into a serious experimental system (e.g., a neural-network-based flight-control system test-flown on a manned NASA F-15 aircraft) or a commercial product (e.g., Sharp’s Logi-cook).

However, the general question of how to make sure that the NN-based adaptive control system performs as expected in all cases has not yet been addressed satisfactorily. While theory and concepts of adaptive systems and intelligent control have been studied in depth over the past decade or so, only very little attention has been paid to the issue of validating the correctness and safety of their operation. All safety-related software applications require careful verification and validation (V&V) of the software components, ranging from extended testing to full-fledged certification procedures (e.g., DO178-B). The adaptive nature of neural networks requires a significantly different approach to verification and validation than used for traditional software, since dynamic adaptation of parameters, iterative numerical algorithms, and complex control architectures renders traditional approaches to V&V impracticable. Many prototypical/experimental application of neural networks in safety-related areas have demonstrated superior behavior and practical usefulness. Unless, however, methods and techniques have been developed which are capable of assuring the correctness and performance of a neural-network based system, NN applicability in safety-critical areas is substantially limited.

The purpose of the workshop is to bring together researchers and users of learning and adaptive systems and control systems in order to create a forum for discussing recent advances in verification, validation, and testing of learning systems, to understand better the practical requirements for developing and deploying neuro-adaptive, and to inspire research on new methods and techniques for verification, validation, and testing.

Topics of interest include but are not limited to:

• applications of learing and adaptive methods and NNs in safety-critical areas and experience/lessons learned.
• applications of collaborative filtering problems, node modeling for belief networks and dependency networks, sequential decision making tasks, diagnosis problems, autonomous systems, robotics, and security, etc.
• techniques, tools, and methods to assess and guarantee the performance of a NN, e.g., statistical (Bayesian) methods, rule extraction with subsequent V&V, methods for convergence/stability analysis, dynamic monitoring of the NN behavior, etc.,
• V&V techniques that are specifically suitable for on-line trained and adaptive systems
• software development, V&V, and certification processes for learning and adaptive systems.

Program