Myeongjin Kang

Mr. Myeongjin Kang


Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
3 January 2022 in Electronics
Verified
Journal Article
Open access
PDF file
JCJuneseo Chang
Juneseo Chang
MKMyeongjin Kang
Myeongjin Kang
Daejin ParkDaejin Park
Daejin Park
Smart homes assist users by providing convenient services from activity classification with the help of machine learning (ML) technology. However, most of the conventional high-performance ML algorithms require relatively high power consumption and memory usage due to their complex structure. Moreover, previous studies on lightweight ML/DL models for human activity classification still require relatively high resources for extremely resource-limited embedded systems; thus, they are inapplicable for smart homes’ embedded system environments. Therefore, in this study, we propose a low-power, memory-efficient, high-speed ML algorithm for smart home activity data classification suitable for an extremely resource-constrained environment. We propose a method for comprehending smart home activity data as image data, hence using the MNIST dataset as a substitute for real-world activity data. The proposed ML algorithm consists of three parts: data preprocessing, training, and classification. In data preprocessing, training data of the same label are grouped into further detailed clusters. The training process generates hyperplanes by accumulating and thresholding from each cluster of preprocessed data. Finally, the classification process classifies input data by calculating the similarity between the input data and each hyperplane using the bitwise-operation-based error function. We verified our algorithm on ‘Raspberry Pi 3’ and ‘STM32 Discovery board’ embedded systems by loading trained hyperplanes and performing classification on 1000 training data. Compared to a linear support vector machine implemented from Tensorflow Lite, the proposed algorithm improved memory usage to 15.41%, power consumption to 41.7%, performance up to 50.4%, and power per accuracy to 39.2%. Moreover, compared to a convolutional neural network model, the proposed model improved memory usage to 15.41%, power consumption to 61.17%, performance to 57.6%, and power per accuracy to 55.4%.

16 August 2021 in Sensors
Verified
Journal Article
Open access
PDF file
MKMyeongjin Kang
Myeongjin Kang
Daejin ParkDaejin Park
Daejin Park
Embedded systems typically operate in harsh environments, such as where there is external shock, insufficient power, or an obsolete sensor after the replacement cycle. Despite these harsh environments, embedded systems require data integrity for accurate operation. Unintended data changes can cause a serious error in reduced instruction set computer (RISC)-based small embedded systems. For instance, if communication is performed on an edge, where there is insufficient power supply, the peak threshold is not reached, resulting in data transmission failure or incorrect data transmission. To ensure data integrity, we use an error-correcting code (ECC), which can detect and correct errors. The ECC parity bit and data are stored together using additional ECC memory, and the original data are extracted through the ECC decoding process. The process of extracting the original data is executed in the instruction fetch stage, where a bottleneck appears in the RISC-based structure. When the ECC decoding process is executed in the bottleneck, the instruction fetch stage increases the instruction fetch time and significantly reduces the overall performance. In this study, we attempt to minimize the effect of ECC on the transmission speed by executing the ECC decoding process in parallel to improve speed by degrading the bottleneck. To evaluate the performance of a parallelized ECC decoding block, we applied the proposed method to the tiny processing unit (TPU) with a RISC-based von Neumann structure and compared memory usage, speed, and reliability according to different transmission success rates in each model. The experiment was conducted using a benchmark that repeatedly executed several 3*3 matrix calculations, and reliability improvement was compared by corrupting the stored random date to confirm the reliability of the transmission success rate. As a result, in the proposed model, using the additional parity bits for parallel processing, memory usage increased by 10 bits per instruction, reducing the data rate from 80 to 61%. However, it showed an improvement in overall reliability and a 7% increase in speed.

7 December 2025 in Sensors
Verified
Journal Article
Open access
PDF file
MKMyeongjin Kang
Myeongjin Kang
Daejin ParkDaejin Park
Daejin Park
Edge inference systems must sustain real-time performance under dynamic environments such as sensor noise, illumination change, and new object classes. Conventional edge devices deploy static offline-trained models, causing accuracy degradation when the input distribution drifts. This study proposes a runtime-robust edge inference framework that enables continuous adaptation without interrupting execution. The edge device partitions its memory into active and adaptive regions, applying task-specific masked updates generated by a server-side FPGA. The FPGA performs layer-wise importance analysis, partial retraining, and adaptive mask generation using dynamic partial reconfiguration (DPR) to minimize reconfiguration delay. Experiments on MNIST, CIFAR-10, and Tiny ImageNet show that the proposed method reduces adaptation latency by up to 1.3× compared with GPU full retraining while cutting the communication cost to 28% of full model transmission. These results demonstrate that combining masking-based selective updates with FPGA DPR acceleration achieves real-time adaptability, low latency, and communication-efficient learning in cloud–edge collaborative environments.