Discovering Natural Laws By Machine

Analysing experimental data is one of important aspects in scientific discovery. We created a powerful algorithm called Reduction and built a Reduction-based discovery system; DISCOVER 2.1. By Reduction, a complex law with multiple variables involved can be discovered by reducing it to a binary law involving only two variables. DISCOVER 2.1, was developed with a flexible knowledge base, an ability of dealing with imperfect data and a proper supporting environment for utility. An analytical comparision indicates that it demonstrates some advantages over the current discovery systems. These include a power of discovering complicated laws without hints, an efficient depth-first search without backtracking and a great potential of utility as a computer-aided data analyser. A clear summary also points out the existing problems and possible solutions, which lead to an out line of our future research.