http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Quantum Approximate Support Vector Machine for NISQ Application
Siheon Park(박시헌),Daniel K. Park(박경덕),June-Koo Rhee(이준구) 대한전자공학회 2021 대한전자공학회 학술대회 Vol.2021 No.6
The previously reported qSVM involves with costly subroutines, which is not suitable for noisy intermediate-scale quantum devices (NISQ). Although Hadamard-test Classifier (HTC) and SWAP-test Classifier (STC) were developed to overcome this problem, they cannot be optimized against complex training samples, performing worse that qSVM. In this work, we propose quantum-classical hybrid variational algorithm for NISQ machines. SVM is transformed to unconstraint problem in this method, which is much more efficient than SVM in terms of number of parameters, required space, and computational complexity. We also present quantum circuits efficient execution fo the algorithm. Finally, we provide numerical analysis with Iris dataset to explain performance with respect to the number of parameters used.
Siheon Lee,Sungkyun Hong,Seungji Kim,Dayoung Oh,Sangkyung Choen,Mincheol CHOI,Junghee YOON 대한수의학회 2020 Journal of Veterinary Science Vol.21 No.4
Background: Quantitative evaluation of renal cortical echogenicity (RCE) has been tried and developed in human and veterinary medicine. Objectives: The objective of this study was to propose a method for evaluating RCE quantitatively and intuitively, and to determine associations between ultrasonographic renal structural distinction and estimated glomerular filtration rate (eGFR) in canine chronic kidney disease (CKD). Methods: Data were collected on 63 dogs, including 27 with normal kidney function and 36 CKD patients. Symmetric dimethylarginine and creatinine concentrations were measured for calculating eGFR. RCE was evaluated as 3 grades on ultrasonography images according to the distinction between the renal cortex and outer medulla. The RCE grade of each kidney was measured. Results: There was a significant difference in eGFR between the group normal and CKD (p < 0.001). As mean of RCE grades (the mean values of each right and left kidney's RCE grade) increases, the proportion of group CKD among the patients in each grade increases (p < 0.001). Also, severity of RCE (classified as "high" if any right or left kidney evaluated as RCE grade 3, "low" otherwise) and eGFR is good indicator for predicting group CKD (p < 0.001). Conclusions: The degree of distinction between the renal cortex and the outer medulla is closely related to renal function including eGFR and the RCE grade defined in this study can be used as a method of objectively evaluating RCE.
Toward an automatic diagnostic scanning robotic system using ultrasound images
Siheon An,Seong Young Ko 제어로봇시스템학회 2022 제어로봇시스템학회 국제학술대회 논문집 Vol.2022 No.11
Ultrasonography is widely used for the diagnosis of various diseases. To obtain a clear view of the ultrasound image, an examiner needs to press the ultrasound (US) probe with a constant press and move smoothly and continuously. To mitigate the workload and prepare the remote treatment, there has been a demand for an automatic diagnostic scanning robotic system (ADSRS) using US images. For the complete ADSRS, the target localization, path planning, robot control, and US image analysis need to be addressed. This paper explains mainly the robot control part as the preliminary research toward ADSRS. A simple path was selected within a predefined inspection area as the preliminary work and the selected path was converted from the camera coordinates into the robot coordinates. Both the position and orientation of the US probe were used as the desired path. For the robot control, hybrid force-positon control was implemented to guarantee the quality of US images and the safety of patients. For the position control, the built-in Cartesian impedance control and null space control were utilized to control the 7-degree-of-freedom (7-DOF) manipulator. For the force control, the basic proportional control was implemented, and it was integrated with the aforementioned position control system. The experimental evaluation was conducted by using a 7-DOF manipulator with a joint torque sensor at each joint after attaching a US probe and a 3-dimensional (3-D) camera to an end-effector. The simple path over the surface of the skin of the patient was generated successfully, and the desired force during movement was set to 5N. The error is measured to 0.3144±0.4853 N. The proposed system shows sufficient performance in the robot control part and the path planning part.