Active Cleaning of Label Noise Applied to ImageNet
Lawrence O. Hall
Distinguished Professor
IEEE/AAAS/IAPR Fellow
Winner of Norbert Wiener Prize
University of South Florida, USA
Abstract: Mislabeled examples in training data can negatively affect the performance of supervised classifiers. We present an approach to remove any mislabeled examples in the dataset by selecting suspicious examples as targets for inspection. We show that the large margin and soft margin principles used in support vector machines (SVM) have the characteristic of capturing the mislabeled examples as support vectors. Experimental results on two character recognition datasets show that oneclass and two-class SVMs are able to capture around 85% and 99% of label noise examples, respectively, as their support vectors. A new method that iteratively builds two-class SVM classifiers on the non-support vector examples from training data followed by an manual expert verification of the support vectors based on their classification score to identify any mislabeled examples is discussed. We show that this method reduces the number of examples to be reviewed through experimental results on four data sets. Finally, we show that this approach finds mislabeled examples in the very large ImageNet database with many fewer examples being examined than a current ad hoc procedure. We argue that by (re-)examining the labels of selective support vectors, most noise can be removed. This can be quite advantageous when rapidly building a labeled dataset such as ImageNet.
Bio: Lawrence O. Hall is a Distinguished University Professor in the Department of Computer Science and Engineering at University of South Florida. In the Fall of 2015 he was the Melchor Visiting Professor in the Department of Computer Science and Engineering and Distinguished Fellow of the Notre Dame Institute for Advanced Study at the University of Notre Dame. He received his Ph.D. in Computer Science from the Florida State University in 1986 and a B.S. in Applied Mathematics from the Florida Institute of Technology in 1980. He is a fellow of the IEEE. He is a fellow of the AAAS and IAPR. He received the Norbert Wiener award in 2012 from the IEEE SMC Society. His research interests lie in distributed machine learning, extreme data mining, bioinformatics, pattern recognition and integrating AI into image processing. The exploitation of imprecision with the use of fuzzy logic in pattern recognition, AI and learning is a research theme. He has authored or coauthored over 80 publications in journals, as well as many conference papers and book chapters. He has received over $3.5M in research funding from agencies such as the National Science Foundation, National Institutes of Health, Department of Energy, NASA, etc.
He received the IEEE SMC Society Outstanding contribution award in 2008. He received an Outstanding Research achievement award from the Univ. of South Florida in 2004. A past president of NAFIPS. The former vice president for membership of the SMC society. He was the President of the IEEE Systems, Man and Cybernetics society for 2006-7. He was the Editor-In-Chief of the IEEE Transactions on Systems, Man and Cybernetics, Part B, 2002-05. He served as the first Vice President for Publications of the IEEE Biometrics Council. He is currently on the IEEE PSPB and chairs its Strategic Planning Committee. He is a member of IEEE PRAC. Also, Editorial board of IEEE Access, associate editor for IEEE Transactions on Fuzzy Systems, International Journal of Intelligent Data Analysis, the International Journal of Pattern Recognition and Artificial Intelligence and International Journal of Approximate Reasoning.