Tell me, please, what are the ways of generating reliable hypotheses, as well as their advantages / disadvantages? What kind of Russian-language literature would you recommend reading on this topic?

  • Well, the questions for you! Google has so far trained only the neural network [cats to recognize] [1], and you already want to put forward theories using AI. [1]: lenta.ru/news/2012/06/27/ggl - Salivan
  • eight
    @Asen, before writing nonsense, at least google the phrase “JSM Method.” - drdaeman September
  • aiportal.ru/downloads/books/… - igumnov

2 answers 2

Found at least 2 methods:

  • GUHA method
  • JSM method

Literature :

Gayek P., Gavranek T. Automatic hypothesis formation. Mathematical foundations of general theory (GUHA method)

JSM-method of automatic hypothesis generation. Logical and epistemological grounds

Automatic hypothesis generation in intelligent systems

Reliable and plausible output in intelligent systems

Among the shortcomings of the JSM-method one can note the inapplicability in the conditions of contradictions in the training examples, as well as

The JSM method that does not use statistical criteria can be “misled” by noise data.

    I do not think that the JSM method will help you in proposing "reliable" hypotheses. For a hypothesis to become reliable, it must be formally proved; but this is no longer a hypothesis, but a theorem. And with regards to the hypotheses, I think in most cases the neural network or the support vector method or the selection tree will fit, if you want to see the hypothesis itself.