Abstract—A column reduction technique for an in-memory machine-learning classifier in 6T SRAM cells is discussed in this paper, based on an error-tolerant boosting algorithm (a.k.a., error-adaptive classifier boosting, EACB). The proposed technique is mainly applied to the in-memory machine-learning classifier system wherein the weight of the linear model is restricted to 1 bit applicable for standard 6T SRAM cells, employing the EACB algorithm to recognize downsampled handwritten digits. First, the number of columns of the boosted classifier is pruned. Second, three methods: greedy search, fast version of greedy search, and worst-care optimization, are discussed and implemented. Finally, the reduction effects of the proposed methods are compared. The simulation results show that besides the 11.50% column reduction from pruning, the proposed methods can further reduce 3.23%, 5.14%, and 5.49% of the column number on average, respectively, with a similar accuracy to ensure that the corresponding part of the model can be reduced to achieve better energy saving.
Index Terms—In-memory, machine learning, column reduction, error-adaptive classifier boosting.
J. Xi is with the Department of Computer Science and Engineering, Fukuoka Institute of Technology, Fukuoka, Japan and the School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing, China (e-mail: firstname.lastname@example.org).
Cite: Jiazhen Xi and Hiroyuki Yamauchi, "A Column Reduction Technique for an In-Memory Machine-Learning Classifier," International Journal of Machine Learning and Computing vol. 8, no. 2, pp. 127-132, 2018.