• ETRI Journal
• /
• 제45권2호
• /
• pp.318-328
• /
• 2023

Recently, embedded systems, such as mobile platforms, have multiple processing units that can operate in parallel, such as centralized processing units (CPUs) and neural processing units (NPUs). We can use deep-learning compilers to generate machine code optimized for these embedded systems from a deep neural network (DNN). However, the deep-learning compilers proposed so far generate codes that sequentially execute DNN operators on a single processing unit or parallel codes for graphic processing units (GPUs). In this study, we propose PartitionTuner, an operator scheduler for deep-learning compilers that supports multiple heterogeneous PUs including CPUs and NPUs. PartitionTuner can generate an operator-scheduling plan that uses all available PUs simultaneously to minimize overall DNN inference time. Operator scheduling is based on the analysis of DNN architecture and the performance profiles of individual and group operators measured on heterogeneous processing units. By the experiments for seven DNNs, PartitionTuner generates scheduling plans that perform 5.03% better than a static type-based operator-scheduling technique for SqueezeNet. In addition, PartitionTuner outperforms recent profiling-based operator-scheduling techniques for ResNet50, ResNet18, and SqueezeNet by 7.18%, 5.36%, and 2.73%, respectively.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1988년도 전기.전자공학 학술대회 논문집
• /
• pp.29-32
• /
• 1988

A new set of the units of demi-syllables is presented for Korean speech synthesis. The performance of the set of demi-syllable units is compared with that of the set of syllable units in the aspects of the quality of synthesized speech using each set of the units and the size of the computer memory which each set of units occupies. The set of demi-syllable units achieves comparable speech quality and occupies smaller memory size than the set of syllable units.

• 대한의용생체공학회:학술대회논문집
• /
• 대한의용생체공학회 1996년도 춘계학술대회
• /
• pp.315-319
• /
• 1996

We have developed a prototype patient monitoring system including module-based bedside units, interbed network, and central stations. A bedside unit consists of a color monitor and a main CPU unit with peripherals including a module controller. It can also include up to 3 module cases and 21 different modules. In addition to the 3-channel recorder module, six different physiological parameters of ECG, respiration, invasive blood pressure, noninvasive blood pressure, body temperature, and arterial pulse oximetry with plethysmogaph are provided as parameter modules. Modules and a module controller communicate with up to 1Mbps data rate through an intrabed network based on RS-485 and HDLC protocol. Bedside units can display up to 12 channels of waveforms with any related numeric informations simultaneously. At the same time, it communicates with other bedside units and central stations through interbed network based on 10Mbps Ethernet and TCP/IP protocol. Software far bedside units and central stations fully utilizes gaphical user interface techniques and all functions are controlled by a rotate/push button on bedside unit and a mouse on central station. The entire system satisfies the requirements of AAMI and ANSI standards in terms of electrical safety and performances. In order to accommodate more advanced data management capabilities such as 24-hour full disclosure, we are developing a relational database server dedicated to the patient monitoring system. We are also developing a clinical workstation with which physicians can review and examine the data from patients through various kinds of computer networks far diagnosis and report generation. Portable bedside units with LCD display and wired or wireless data communication capability will be developed in the near future. New parameter modules including cardiac output, capnograph, and other gas analysis functions will be added.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2005년도 춘계학술대회 논문집
• /
• pp.1307-1310
• /
• 2005

ATM is element that reliability, stability is important in relation with customer by financial agency used device. To embody high reliability and stability, reliability estimation technology development is important. When new product is developed, performance and reliability evaluation of product are essential element. In this paper, is treating contents on reliability estimation of stepping motor, BLDC motor and solenoid that is main driving source of note handling units.

• 전자공학회논문지C
• /
• 제35C권11호
• /
• pp.89-98
• /
• 1998

정적 시스템 모델링을 위해 RBF 신경회로망의 은닉 유니트를 자동으로 생성하는 ERAN을 제안한다. ERAN은 관측 데이터의 신규성을 기반으로 새로운 은닉 유니트를 할당하는 RAN의 성능을 개선한 것이다. ERAN의 학습 과정은 새로운 은닉 유니트의 생성과 네트웍 파라미터 학습을 포함한다. 네트웍은 초기에 0개의 은닉 유니트로 시작하여 세 가지의 은닉 유니트 생성 판별기준을 만족할 경우에만 새로운 은닉 유니트를 생성시킨다. 네트웍의 파라미터는 LMS 알고리즘을 이용하여 조정한다. 제안한 ERAN의 성능은 순차 학습 및 랜덤 학습을 갖는 비선형 정적 시스템 모델링 문제에 대하여 RAN의 결과와 성능을 비교한다. 두 실험에 대하여 ERAN은 RAN 보다 적은 은닉 유니트를 가지고 정확성이 더 우수한 RBF 신경회로망을 구현할 수 있음을 보인다.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2005년도 Proceedings of ISRS 2005
• /
• pp.659-662
• /
• 2005

Recently, the need for LBS (Location Based Services) is increasing with the rapid growth of the location measurement units, mobile devices, and communication technologies. Especially, wireless carriers are concentrating on LBS since it is regarded as so-called 'killer application' among wireless data services. Although LBS give us convenience and useful information, its use also raises privacy issues. There are quite possibilities that the people's locations are tracked by location measurement units while people do not recognize the existence of the units. To protect a person's location information, lawful and technical aspects should be considered. In this paper, we explain issues, regulations, standards, and platforms related to the protection of the location information. Finally, we suggest the architecture of a platform complying with the regulations and standards. It considers various issues not covered in other studies.

• 전자공학회논문지B
• /
• 제31B권4호
• /
• pp.129-137
• /
• 1994

In this paper, a new texture feature extraction method for texture image classification is proposed. The proposed method is a modified texture spectrum method. It uses local weighting averaged texture unit, that is, the neighbor pixels are weithted and averaged in 4-direction and the calculated values are compared with center pixel to find texture units. The proposed method has only 81 texture units and these units are really good features for texture classification. The proposed method is applied to vegetable images and Blodatz album images and compared with several conventional methods for the feature extraction time and the recognition rate.

• 한국재료학회:학술대회논문집
• /
• 한국재료학회 2009년도 춘계학술발표대회
• /
• pp.15.2-15.2
• /
• 2009

Manufacturing of printed electronics using printing technology has begun to get into the hot issue in many ways due to the low cost effectiveness to existing semi-conductor process. This technology with both low cost and high productivity, can be applied in the production of organic thin film transistor (OTFT), solar cell, radio frequency identification (RFID) tag, printed battery, E-paper, touch screen panel, black matrix for liquid crystal display (LCD), flexible display, and so forth. The emerging technology to manufacture the products in mass production is roll-to-roll printing technology which is a manufacturing method by printings of multi-layered patterns composed of semi-conductive, dielectric and conductive layers. In contrary to the conventional printing machines in which printing precision is about $50~100{\mu}m$, the printing machines for printed electronics should have a precision under $30{\mu}m$. In general, in order to implement printed electronics, narrow width and gap printing, register of multi-layer printing by several printing units, and printing accuracy of under $30{\mu}m$ are all required. We developed the roll-to-roll printing equipment used for printed electronics, which is composed of un-winder, re-winder, tension measurement system, feeding units, dancer systems, guide unit, printing unit, vision system, dryer units, and various auxiliary devices. The equipment is designed based on cantilever type in which all rollers except printing ones have cantilever types, which could give more accurate machine precision as well as convenience for changing rollers and observing the process.

• 전자공학회논문지
• /
• 제51권11호
• /
• pp.193-201
• /
• 2014

본 논문에서는 얼굴영상에 나타난 사람의 표정을 인식하기 위해 얼굴검출, 얼굴정렬, 얼굴단위 추출, 그리고 AdaBoost를 이용한 학습 방법과 효과적인 인식방법을 제안한다. 입력영상에서 얼굴 영역을 찾기 위해서 얼굴검출을 수행하고, 검출된 얼굴영상에 대하여 학습된 얼굴모델과 정렬(Face Alignment)을 수행한 후, 얼굴의 표정을 나타내는 단위요소(Facial Units)들을 추출한다. 본 논문에서 제안하는 얼굴 단위요소들을 표정을 표현하기 위한 기본적인 액션유닛(AU, Action Units)의 하위집합으로 눈썹, 눈, 코, 입 부분으로 나눠지며, 이러한 액션유닛에 대하여 AdaBoost 학습을 수행하여 표정을 인식한다. 얼굴유닛은 얼굴표정을 더욱 효율적으로 표현할 수 있고 학습 및 테스트에서 동작하는 시간을 줄여주기 때문에 실시간 응용분야에 적용하기 적합하다. 실험결과, 제안하는 표정인식 시스템은 실시간 환경에서 90% 이상의 우수한 성능을 보여준다.

• ETRI Journal
• /
• 제45권4호
• /
• pp.724-724
• /
• 2023