A Self-Training System That Learns Through Experimentation

摘要: 1 Motivation The knowledge in computer programs is usually coded into the software itself or stored a linked database. program's developer provides system with he, his client, thinks needed to fulfil task it designed for. Commonly gives user possibility add more system, for example by adding new dataset system's can be described as predefined and fixed not able change without external help. That means depended on developer's user's knowledge. For some applications possible reasonable predefine This case if needs certain either unknown at time of its development too manifold complex. Often developers code additional may useful. In contrast this kind software, tools are adapt themselves their task. Such programmed increase through training. training supervised learn autonomously. both cases data has generated before phase begins. Usually process carried out human takes time. Not only design but also learning consuming. Flexible, adaptive could useful user-friendly were reduce expend . 2 Aim study presents an approach where such learns idea have that generate train derived from data. exemplary here form basis tool used assist designers during early stages product design. As any situated dynamic (Gero Kannengiesser 2004), systems developed designer should adjust changing conditions use. specific domain demonstration recognition substitution hand-drawn geometrical shapes. A like transform conceptual sketch refined CAD representation.