Huang S.-AHsieh Y.-YCHIA-HSIANG YANG2022-04-252022-04-252021https://www.scopus.com/inward/record.uri?eid=2-s2.0-85113345667&doi=10.1109%2fAICAS51828.2021.9458569&partnerID=40&md5=d5f3a924da880ae2243dc3942d031be4https://scholars.lib.ntu.edu.tw/handle/123456789/606981This paper presents an optimized support vector machine (SVM) training processor employing the alternative direction method of multipliers (ADMM) optimizer. Low-rank approximation is exploited to reduce the dimension of the kernel matrix by employing the Nystr?m method. Verified in four datasets, the proposed ADMM-based training processor with rank approximation reduces 32 of matrix dimension with only 2% drop in inference accuracy. Compared to the conventional sequential minimal optimization (SMO) algorithm, the ADMM-based training algorithm is able to achieve a 9.8 10^{7} shorter latency for training 2048 samples. Hardware optimization techniques, including pre-computation and memory sharing, are proposed to reduce the computational complexity by 62% and the memory usage by 60%. As a proof of concept, an epileptic seizure detector is designed to demonstrate the effectiveness of the proposed optimization techniques. The chip achieves a 153,310 higher energy efficiency and a 364 higher throughput-to-area ratio for SVM training than a high-end CPU. This work provides a promising solution for edge devices which require low-power and real-time training. ? 2021 IEEE.alternative direction method of multipliers (ADMM)hardware-efficient realizationon-line trainingrank approximationSupport vector machine (SVM)Approximation theoryEdge computingEnergy efficiencyIntegrated circuit designMatrix algebraOptimizationDesign optimizationHardware optimizationLow rank approximationsMethod of multipliersOptimization techniquesReal-time trainingSequential minimal optimization algorithmsTraining algorithmsSupport vector machines[SDGs]SDG7conference paper10.1109/AICAS51828.2021.94585692-s2.0-85113345667