A Generic Neural Network Framework Using Design Patterns

摘要: Designing object-oriented software is hard, and designing reusable even harder. This task more daunting for a developer of computational intelligence applications, as optimising one design objective tends to make others inefficient or impossible. Classic examples in computer science include ‘storage vs. time’ ‘simplicity flexibility.’ Neural network requirements are by their very nature tightly coupled – required change area an existing application have severe effects other areas, making the impossible inefficient. Often this situation leads major redesign system many cases completely rewritten application. Many commercial open-source packages do exist, but these cannot always be extended support input from fields due proprietary reasons failing fully take all into consideration. Design patterns out writing that modular, extensible efficient well easy read understand. The essence pattern avoid repeatedly solving same problem scratch reusing solution solves core problem. template has wellunderstood prerequisites, structure, properties, behaviour consequences. CILib framework allows developers develop new applications quickly efficiently. Flexibility, reusability clear separation between components maximised through use patterns. Reliability also ensured open source thus people collaborate ensure designed error free. dissertation discusses implementation generic neural users design, implement any possible models algorithms such way they can reuse reused algorithm rest framework, external applications. achieved using framework.