Theory formation by heuristic search

摘要: Machine learning can be categorized along many dimensions, an important one of which is 'degree human guidance or forethought'. This continuum stretches from rote learning, through carefully-guided concept-formation by observation, out toward independent theory formation. Six years ago, the am program was constructed as experiment in this latter kind discovery. Its source power a large body heuristics, rules guided it fruitful topics investigation, profitable experiments to perform, plausible hypotheses and definitions. Since that time, we have gained deeper insight into nature heuristics process forming extending theories empirically. 'The Nature Heuristics I' paper presented theoretical basis for work, with emphasis on how relate each other. presents our accretion model formation, gives examples its use producing new discoveries various fields. These are drawn runs called eurisko, successor am, embodies uses corpus guide behavior. demands ability discover periodically, well domain objects operators, some illustrate well. III' describes architecture eurisko program, conclusions made