Characterizing Driver Stress Using Physiological and Operational Data from Real-World Electric Vehicle Driving Experiment

Kim, Seyun,Rhee, Wonjong,Choi, Daeyoung,Jang, Young Jae,Yoon, Yoonjin Springer-Verlag 2018 International journal of automotive technology Vol.19 No.5

Partitioning Compute Units in CNN Acceleration for Statistical Memory Traffic Shaping

Jung, Daejin,Lee, Sunjung,Rhee, Wonjong,Ahn, Jung Ho IEEE 2018 IEEE computer architecture letters Vol.17 No.1

<P>Convolutional Neural Networks (CNNs) have become the default choice for processing visual information, and the design complexity of CNNs has been steadily increasing to improve accuracy. To cope with the massive amount of computation needed for such complex CNNs, the latest solutions utilize blocking of an image over the available dimensions (e.g., horizontal, vertical, channel, and kernel) and batching of multiple input images to improve data reuse in the memory hierarchy. While there has been a large collection of works on maximizing data reuse, only a few studies have focused on the memory bottleneck problem caused by limited bandwidth. Bandwidth bottleneck can easily occur in CNN acceleration as CNN layers have different sizes with varying computation needs and as batching is typically performed over each layer of CNN for an ideal data reuse. In this case, the data transfer demand for a layer can be relatively low or high compared to the computation requirement of the layer, and therefore temporal fluctuations in memory access can be induced eventually causing bandwidth problems. In this paper, we first show that there exists a high degree of fluctuation in memory access to computation ratio depending on CNN layers and functions in the layer being processed by the compute units (cores), where the compute units are tightly synchronized to maximize data reuse. Then we propose a strategy of partitioning the compute units where the cores within each partition process a batch of input data in a synchronous manner to maximize data reuse but different partitions run asynchronously. Because the partitions stay asynchronous and typically process different CNN layers at any given moment, the memory access traffic sizes of the partitions become statistically shuffled. Thus, the partitioning of compute units and asynchronous use of them make the total memory access traffic size be smoothened over time, and the degree of partitioning determines a tradeoff between data reuse efficiency and memory bandwidth utilization efficiency. We call this smoothing statistical memory traffic shaping, and we show that it can lead to 8.0 percent of performance gain on a commercial 64-core processor when running ResNet-50.</P>

강화학습의 학습속도 및 성능 향상을 위한 메타러닝과 초모수 최적화

조형헌(Hyunghun Cho),이원종(Wonjong Rhee) 한국정보과학회 2019 정보과학회지 Vol.37 No.11

Static and Dynamic Yaw Misalignments of Wind Turbines and Machine Learning Based Correction Methods Using LiDAR Data

Choi, Daeyoung,Shin, Won,Ko, Kyungnam,Rhee, Wonjong IEEE 2019 IEEE transactions on sustainable energy Vol.10 No.2

<P>A yaw misalignment can be static or dynamic depending on its variation over time. For static, a few popular correction methods exist. LiDAR is one of the promising solutions because it can provide accurate wind measurements compared to a vane sensor and because it is cost effective when used for a limited time. We extend the LiDAR method to dynamic yaw misalignment correction by modeling the misalignment error's dependence on wind direction, wind speed, and rotor speed. An analytical framework is developed, and machine learning algorithms are trained to estimate the LiDAR's wind direction using SCADA data only. In this way, dynamic errors within SCADA data can be mitigated even after LiDAR is removed. Three machine learning algorithms, linear regression, random forests, and gradient boosting, are investigated. To evaluate the algorithms, SCADA and LiDAR data were collected at two wind farm sites in South Korea. The analysis shows that machine learning algorithms are capable of mitigating both static and dynamic yaw misalignments. While error reduction of only 22.6% was achieved with a static method, error reduction of 44.4% was achieved with a machine learning method. The validity was double checked by investigating turbine-to-turbine transferability of the dynamic correction model.</P>

Assuring explainability on demand response targeting via credit scoring

Lee, Kyungeun,Lee, Hyesu,Lee, Hyoseop,Yoon, Yoonjin,Lee, Eunjung,Rhee, Wonjong Elsevier 2018 ENERGY Vol.161 No.-

<P><B>Abstract</B></P> <P>As data-driven innovation becomes a main trend in the energy sector, explainability of data-driven actions is becoming a major fairness issue for the residential applications, and it is expected to become a requirement for regulatory compliance. Explainability, however, often demands a sacrifice in prediction performance and affects the effectiveness of data-driven actions. In this study, we consider data-driven customer targeting in an incentive-based residential demand response program, and investigate the explainability-performance tradeoff when using simple-rule based, machine learning, and credit scoring methods. Credit scoring, that has been a popular solution in the finance discipline for over 60 years, is a scorecard based modeling method that can surely provide explainability. We first provide the detailed steps of applying credit scoring to the demand response problem. Then, we use a dataset of 14,525 households obtained from a real demand response program and analyze two prediction problems – participation prediction and behavior change prediction. The results show that credit scoring can achieve a comparable performance as the best-performing machine learning methods while providing full explainability. Our results suggest that credit scoring can be a promising explainability option for broader energy sector problems.</P> <P><B>Highlights</B></P> <P> <UL> <LI> A quantitative analysis of data-driven targeting in residential DR. </LI> <LI> Explainability of data-driven actions and its relation to fairness. </LI> <LI> Details of implementing credit scoring, which has good explainability, for DR. </LI> <LI> A case study of incentive DR, where the DR was operated through a smartphone app. </LI> <LI> Credit scoring can achieve a comparable performance as machine learning methods. </LI> </UL> </P>

Machine Learning Prediction for the Recurrence After Electrical Cardioversion of Patients With Persistent Atrial Fibrillation

Soonil Kwon,Eunjung Lee,Hojin Ju,Hyo-Jeong Ahn,So-Ryoung Lee,Eue-Keun Choi,Jangwon Suh,Seil Oh,Wonjong Rhee 대한심장학회 2023 Korean Circulation Journal Vol.53 No.10

Background and Objectives: There is limited evidence regarding machine-learning prediction for the recurrence of atrial fibrillation (AF) after electrical cardioversion (ECV). This study aimed to predict the recurrence of AF after ECV using machine learning of clinical features and electrocardiograms (ECGs) in persistent AF patients. Methods: We analyzed patients who underwent successful ECV for persistent AF. Machine learning was designed to predict patients with 1-month recurrence. Individual 12-lead ECGs were collected before and after ECV. Various clinical features were collected and trained the extreme gradient boost (XGBoost)-based model. Ten-fold cross-validation was used to evaluate the performance of the model. The performance was compared to the C-statistics of the selected clinical features. Results: Among 718 patients (mean age 63.5±9.3 years, men 78.8%), AF recurred in 435 (60.6%) patients after 1 month. With the XGBoost-based model, the areas under the receiver operating characteristic curves (AUROCs) were 0.57, 0.60, and 0.63 if the model was trained by clinical features, ECGs, and both (the final model), respectively. For the final model, the sensitivity, specificity, and F1-score were 84.7%, 28.2%, and 0.73, respectively. Although the AF duration showed the best predictive performance (AUROC, 0.58) among the clinical features, it was significantly lower than that of the final machine-learning model (p<0.001). Additional training of extended monitoring data of 15-minute single-lead ECG and photoplethysmography in available patients (n=261) did not significantly improve the model’s performance. Conclusions: Machine learning showed modest performance in predicting AF recurrence after ECV in persistent AF patients, warranting further validation studies.

확장된 기술수용모델을 이용한 가정용 에너지 수요반응 프로그램 실증분석

정은아(Euna Jung),이경은(Kyungeun Lee),김화영(Hwayoung Kim),정소라(Sora Jeong),이효섭(Hyoseop Lee),서봉원(Bongwon Suh),이원종(Wonjong Rhee) 한국HCI학회 2017 한국HCI학회 논문지 Vol.12 No.4

전력 수요가 증가하고 재생 가능 에너지에 대한 관심이 증폭됨에 따라, 수요를 억제하여 필요한 공급량을 줄일 수 있는 ‘수요반응’ 프로그램에 대한 관심이 증가하고 있다. 본 연구는 가정에 스마트미터를 구비한 국내 사용자들을 대상으로 진행된 에너지 수요반응 실증사업에 대한 실증분석으로, 사전심층 인터뷰, 설문 및 기술수용모델 분석을 통하여 가정 전력 사용자들이 수요반응 프로그램을 받아들이는 데 중요한 요인들을 살펴본다. 수요반응의 목표는 피크시간대에 미션이 발령되면 전력사용량을 평소보다 줄이는 것이며, 실험대상은 스마트미터 구입 경로와 에너지를 절감했을 때 보상받는 방식에 따라 2개의 상이한 집단으로 구성되었다. 집단 A는 주로 IoT플랫폼 서비스에 가입하는 과정에서 마케터와의 대화를 통해 전체 서비스 중 하나인 스마트미터 서비스에 함께 가입하는 경로로 수요반응 프로그램에 유입되었고, 보상으로는 통신비 할인을 받았다. 반면 집단 B는 스마트미터를 자발적으로 구매하거나 에너지 자립 마을 지역주민으로서 지자체 지원을 통해 스마트미터를 지원 받아 프로그램에 유입되었고, 미션 성공에 대한 보상은 사회적 기부를 통해 이루어졌다. 분석 결과 집단 A는 인지된 용이성과 인지된 유용성 외에 인지된 유희성도 포함된 확장된 기술수용모델이 적합함을 알 수 있었고,집단 B는 모델의 적합도가 떨어지기는 하지만 집단 A에 비해 인지된 유용성에 대한 중요도가 높음을 확인할 수 있었다. 이와 같은 결과는 집단 특성에 따른 프로그램 설계방향을 제시하여 향후 수요반응 프로그램을 효과적으로 운영하는 데에 도움을 줄 것으로 보인다. While electricity demand is generally increasing, stably controlling supply is becoming a serious challenge because renewable energies are becoming popular and often their productions are dependent on the weather. The ‘demand response’ programs can be used to complement the problems of renewable energies, and therefore their role is becoming increasingly important. This study provides an analysis of a demand response pilot that was conducted in Korea. The study first focused on questionnaire surveys and in-depth interviews, and the data was used to perform a Technology Acceptance Model (TAM) analysis. The goal of the pilot was to have the residential users reduce their power consumptions when an energy reduction mission is issued during peak load hours. The experimental subjects consisted of two groups with different characteristics. Subjects in group A obtained smart meters as an optional function of IoT platform service provided by a mobile service company, and received a charge deduction as their compensation. Subjects in group B either voluntarily purchased smart meters as individuals or received them by participating in an energy self-sufficient village program that was run by a local government, and were entitled to a donation as their compensation. With the analysis, group A was found to fit the extended technology acceptance model that includes perceived playfulness in addition to perceived ease of use and perceived usefulness. On the contrary, group B failed to fit the model well, but perceived usefulness was found to be relatively more important compared to group A. The results indicate that the residential energy groups’ behavior changes are dependent on each group’s characteristics, and group-specific DR design should be considered to improve the effectiveness of DR.

Extracellular Vesicles-based Acid-responsive Sonotheranostic Agents for Chemo-sonodynamic Cancer Therapy and Photoacoustic Imaging

Thuy Giang NGUYEN CAO,JiHee KANG,WanGyu KIM,SuJin KANG,Hoang Quan TRUONG,WonJong RHEE,ChulHong KIM,YoungTag KO,MinSuk SHIM 한국생물공학회 2022 한국생물공학회 학술대회 Vol.2022 No.9

소규모 전력 소비자 대상 수요자원 거래시장의 필요성 및 시범운영 결과 분석

이은정(Eun-jung Lee),이경은(Kyung-eun Lee),이혜수(Hye-su Lee),이효섭(Hyo-seop Lee),김은철(Eun-cheol Kim),이원종(Wonjong Rhee) 한국통신학회 2017 韓國通信學會論文誌 Vol.42 No.4