0497831 20240103221057.720181210235959.9 85061142387 272 19 s. P cav_un_epca*0497830978-1-53614-429-1A Statistical Review of the MNIST Benchmark Data Problem172-193New YorkNova Science Publishers, Inc.2018LuT.ChaoT.H. MNIST benchmark multivariate Bernoulli mixtures EM algorithm cav_un_auth*0101091 Department of Pattern Recognition 50 Grim Jiří UTIA-B Rozpoznávání obrazu Department of Pattern Recognition RO RO Ústav teorie informace a automatizace AV ČR, v. v. i. cav_un_auth*0101197 Department of Pattern Recognition 50 Somol Petr UTIA-B Rozpoznávání obrazu Department of Pattern Recognition RO RO Ústav teorie informace a automatizace AV ČR, v. v. i. http://library.utia.cas.cz/separaty/2018/RO/grim-0497831.pdf cav_un_auth*0347019 GA17-18407S GA ČR The recognition of MNIST numerals is discussed as a benchmark problem. Applying the probabilistic neural networks to MNIST data we have found that the training and test set have slightly different statistical properties with negative consequences for classifier performance. We assume that the frequently used extension of MNIST training data by distorted patterns improves the recognition accuracy by creating images similar to the atypical test set numerals. In this way the benchmark experiments may be influenced by the external knowledge about the hand-written digits and the comparative value of the benchmark becomes more or less limited to recognition of MNIST numerals. As a more generally applicable benchmark model we propose recognition of artificial binary patterns generated on a chessboard by random moves of the pieces rook and knight. IN 10000 10200 10201 2019 2 RVO:67985556 http://hdl.handle.net/11104/0290648 S 2018 85061142387 SCOPUS PUBMED WOS cav_un_epca*0497830 Advances in Pattern Recognition Research A Statistical Review of the MNIST Benchmark Data Problem Nova Science Publishers, Inc. 2018 New York 172 193 978-1-53614-429-1 340 Lu T. 340 Chao T.H.