• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.1
• /
• pp.147-165
• /
• 2021

In this work, we empirically evaluated the efficiency of the recent EfficientNetB0 model to identify and diagnose malaria parasite infections in blood smears. The dataset used was collected and classified by relevant experts from the Lister Hill National Centre for Biomedical Communications (LHNCBC). We prepared our samples with minimal image transformations as opposed to others, as we focused more on the feature extraction capability of the EfficientNetB0 baseline model. We applied transfer learning to increase the initial feature sets and reduced the training time to train our model. We then fine-tuned it to work with our proposed layers and re-trained the entire model to learn from our prepared dataset. The highest overall accuracy attained from our evaluated results was 94.70% from fifty epochs and followed by 94.68% within just ten. Additional visualization and analysis using the Gradient-weighted Class Activation Mapping (Grad-CAM) algorithm visualized how effectively our fine-tuned EfficientNetB0 detected infections better than other recent state-of-the-art DCNN models. This study, therefore, concludes that when fine-tuned, the recent EfficientNetB0 will generate highly accurate deep learning solutions for the identification of malaria parasites in blood smears without the need for stringent pre-processing, optimization, or data augmentation of images.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.343-347
• /
• 2022

최근 자연어처리 분야에서는 BERT, RoBERTa, 그리고 BART와 같은 사전 학습된 언어 모델 (Pre-trained Language Models, PLM) 기반 미세 조정 학습을 통하여 여러 하위 과업에서 좋은 성능을 거두고 있다. 이는 사전 학습된 언어 모델 및 데이터 집합의 크기, 그리고 모델 구성의 중요성을 보여주며 대규모 사전 학습된 언어 모델이 각광받는 계기가 되었다. 하지만, 거대한 모델의 크기로 인하여 실제 산업에서 쉽게 쓰이기 힘들다는 단점이 명백히 존재함에 따라 최근 매개변수 효율적인 미세 조정 및 Few-Shot 학습 연구가 많은 주목을 받고 있다. 본 논문은 Prompt tuning, Prefix tuning와 프롬프트 기반 미세 조정 (Prompt-based fine-tuning)을 결합한 Few-Shot 학습 연구를 제안한다. 제안한 방법은 미세 조정 ←→ 사전 학습 간의 지식 격차를 줄일 뿐만 아니라 기존의 일반적인 미세 조정 기반 Few-Shot 학습 성능보다 크게 향상됨을 보인다.

• Proceedings of the IEEK Conference
• /
• 2001.09a
• /
• pp.95-98
• /
• 2001

The adaptive line enhancer (ALE) is widely used for enhancing narrowband signals corrupted by broadband noise. In this paper, we propose novel ALE methods to improve the enhancing capability. The proposed methods are motivated by the fact that the output of the ALE is a fine estimate of the desired narrowband signal with the broadband noise component suppressed. The proposed methods preprocess the input signal using ALE filter to regenerate a finer input signal. Thus the proposed ALE is driven by the input signal with higher signal-to-noise ratio (SNR). The analysis and simulation results are presented to demonstrate that the proposed ALE has better performance than conventional ALE´s.

• Current Optics and Photonics
• /
• v.2 no.1
• /
• pp.53-58
• /
• 2018

A novel X-band high speed frequency sweep signal generator based on a tunable optoelectronic oscillator (OEO) incorporating a frequency-swept laser is presented and the theoretical fundamentals of the design are explained. A prototype of the generator with tuning range from 8.8552 GHz to 10.3992 GHz and a fine step about 8 MHz is achieved. The generated radiofrequency signal with a single sideband (SSB) phase noise lower than -100 dBc/Hz@10KHz is experimentally demonstrated within the whole tunable range, without any narrow RF band-pass filters in the loop. And the tuning speed of the frequency sweep signal generator can reach to over 1 GHz/s benefiting from applying a novel dispersion compensation modular instead of several tens of kilometers of optical fiber delay line in the system.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.15 no.4
• /
• pp.437-444
• /
• 2015

A high output power voltage controlled oscillator (VCO) in the U-band was implemented using a 65 nm CMOS process. The proposed VCO used a transmission line to increase output voltage swing and overcome the limitations of CMOS technologies. Two varactor banks were used for fine tuning with a 5% frequency tuning range. The proposed VCO showed small variation in output voltage and operated at 51.55-54.18 GHz. The measured phase noises were -51.53 dBc/Hz, -91.84 dBc/Hz, and -101.07 dBc/Hz at offset frequencies of 10 kHz, 1 MHz, and 10 MHz, respectively, with stable output power. The chip area, including the output buffer, is $0.16{\times}0.16mm^2$ and the maximum output power was -0.11 dBm. The power consumption was 33.4 mW with a supply voltage of 1.2-V. The measured $FOM_P$ was -190.8 dBc/Hz.

• Proceedings of the Korean Institute of Communication Sciences Conference
• /
• 1986.10a
• /
• pp.28-34
• /
• 1986

The frequency synthesizer for Fast Frequency Hopping System musy be capable of a fast tuning with a small step frequency over wide band. The most conventional frequency synthesizer that uses the phase locked loop (PLL) enables the wide band problem but have a poor side of the low resolution and the transient response. In this paper, we have discussed the experimental results of a direct digital frequency synthesizer which can be applicable to the Fast Frequency Hopping System, using digital-to-analoq (D/A)conversion techniques. With this system we can find the merits of a fine resolution and the possibility of a fast tuning leaving the problems of transent frequency.

• Proceedings of the Korea Electromagnetic Engineering Society Conference
• /
• 2001.11a
• /
• pp.257-261
• /
• 2001

This paper presents the implementation of wide-band RFIC amplifier operating from near 50MHz to 3GHz using Tachyonics SiGe HBT foundry. Voltage shunt feedback is used for the flat gain and the broad band impedance matching. Initial design parameters are calculated using the low frequency small signal analysis. Since the HBT model was not available at the design time, discrete tuning board was made for fine tuning in the low frequency range. Fabricated amplifier shows 12dB gain with 1dB fluctuation and PldB reaches 15dBm at 850MHz.

• Proceedings of the KIEE Conference
• /
• 2002.11c
• /
• pp.333-336
• /
• 2002

In many cases, fuzzy modeling has a defect that the design procedure cannot be theoretically justified. To overcome this difficulty, we suggest a new design method for fuzzy model by combining genetic algorithm(GA) and mush set theory. GA, which has the advantages is optimization, and rule base. However, it is some what time consuming, so are introduce rough set theory to the rule reduction procedure. As a result, the decrease of learning time and the considerable rate of rule reduction is achieved without loss of useful information. The preposed algorithm is composed of three stages; First stage is quasi-optimization of fuzzy model using GA(coarse tuning). Next the obtained rule base is reduced by rough set concept(rule reduction). Finally we perform re-optimization of the membership functions by GA(fine tuning). To check the effectiveness of the suggested algorithm, examples for time series prediction are examined.

• ETRI Journal
• /
• v.38 no.5
• /
• pp.879-884
• /
• 2016

In this paper, an ultra-small marker beacon antenna operated in the VHF-band is proposed. This antenna has a modified linear IFA structure with a lumped capacitor and a capacitive plate for frequency tuning and impedance matching. The capacitive plate is directly connected to the end of a linear radiator and is separated from the antenna ground by 1 mm. The main operating frequency is mainly controlled by the size and dielectric constant of the capacitive plate. The lumped capacitor is useful for fine frequency tuning. Using the proposed structure, an ultra-small marker beacon antenna can be realized with a length of 0.04 ${\lambda}_0$.

• Journal of Information Display
• /
• v.13 no.4
• /
• pp.167-177
• /
• 2012

Cold cathode fluorescent lamps (CCFLs) are used to provide lighting for liquid crystal displays (LCDs). This paper presents a set of guidelines for measurement characterization and design to reduce the power consumption of CCFL LCD backlight inverters and panel electronics. The proposed methods aim to reduce the backlight power consumption by fine-tuning a back-light inverter for a specific LCD, using several methods. First, the authors describe their power measurement methodology; and next, they identify different areas for tuning a backlight inverter for a given display. The experiment results showed that power savings can range from 50 to 200mW if the backlight inverter is properly tuned. This paper also proposes an optimized configuration for light-emitting device (LED) panels to reduce power loss by selecting a LED with a specific input voltage and number of cells to help minimize power loss.