We present a new algorithm for finding low-complexity neural networks with high generalization capability. The algorithm searches for a "flat" minimum of the error function. A flat minimum is a large connected region in weight space where the error remains approximately constant. An MDL-based, Bayesian argument suggests that flat minima correspond to "simple" networks and low expected overfitti...

Conifers, a component of beauties of parks and landscapes, are infected with diverse pathogens, especially fungi. In this research, samples were collected from conifers displaying symptoms of disease in parks and landscapes of Rasht in Iran and for the isolation and identification of pathogenic fungi, pieces of infected tissues were placed in PDA and at the next stages in WA, PCA and SNA media ...