A Few English Words Can Help Improve Your Russian.

摘要: This paper reports effects of data set modifications on learning which imply that simplifying a task can make harder, while adding to it easier. These were observed during tests carried out for automatic finite-state automaton (FSA) induction from positive with genetic algorithm (GA), in the state-transition matrices (deterministic) FSAs are directly encoded as variable-length genotypes, and fitness function rewards smallness, ability parse strings approximation given target language size. The GA (general results reported [1]) was successfully applied complete incomplete samples different degrees complexity. For (not NP-hard) is outperformed accuracy speed by approaches based initial construction canonical automaton. developed (NP-hard) compares well existing techniques, although comparison other GA-based techniques difficult, most differ fundamentally terms learnable classes (often required be structurally complete). does outperform achieved comparable approach [3], without requiring negative examples or manual selection.