Open AccessLayer‐wise hint‐based training for knowledge transfer in a teacher‐student framework
Author(s) -
Bae JiHoon,
Yim Junho,
Kim NaeSoo,
Pyo CheolSig,
Kim Junmo
Publication year - 2019
Publication title -
etri journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.295
H-Index - 46
eISSN - 2233-7326
pISSN - 1225-6463
DOI - 10.4218/etrij.2018-0152
Subject(s) - residual neural network , computer science , benchmark (surveying) , mnist database , artificial intelligence , machine learning , residual , transfer of learning , training (meteorology) , layer (electronics) , pattern recognition (psychology) , training set , artificial neural network , algorithm , chemistry , physics , geodesy , organic chemistry , meteorology , geography
We devise a layer‐wise hint training method to improve the existing hint‐based knowledge distillation (KD) training approach, which is employed for knowledge transfer in a teacher‐student framework using a residual network (ResNet). To achieve this objective, the proposed method first iteratively trains the student ResNet and incrementally employs hint‐based information extracted from the pretrained teacher ResNet containing several hint and guided layers. Next, typical softening factor‐based KD training is performed using the previously estimated hint‐based information. We compare the recognition accuracy of the proposed approach with that of KD training without hints, hint‐based KD training, and ResNet‐based layer‐wise pretraining using reliable datasets, including CIFAR‐10, CIFAR‐100, and MNIST. When using the selected multiple hint‐based information items and their layer‐wise transfer in the proposed method, the trained student ResNet more accurately reflects the pretrained teacher ResNet's rich information than the baseline training methods, for all the benchmark datasets we consider in this study.