Data-driven and expectation-driven discovery of empirical laws
暂无分享,去创建一个
Abstract : BACON.5 is a program that discovers empirical laws for summarizing data. The system incorporates four data-driven heuristics for relating numeric terms, recursing to higher levels of description, postulating intrinsic properties such as mass and specific heat, and finding common divisors. BACON.5 also includes expectation-driven strategies for directing search based on discoveries that the program has already made. These include heuristics for expecting similar forms of laws, reducing the amount of data that must be gathered, and taking advantage of the symmetrical form of some laws. BACON.5 has shown its generality by rediscovering a number of laws from the history of physics and chemistry, including Snell's law of refraction, conservation of momentum, and Black's specific heat law.