• Journal of Korea Society of Industrial Information Systems
• /
• v.29 no.3
• /
• pp.1-12
• /
• 2024

In this paper, we fine-tuned KoAlpaca, a large-scale Korean language model, and implemented a blog text generation system utilizing it. Blogs on social media platforms are widely used as a marketing tool for businesses. We constructed training data of positive reviews through emotion analysis and refinement of collected sponsored blog texts and applied QLoRA for the lightweight training of KoAlpaca. QLoRA is a fine-tuning approach that significantly reduces the memory usage required for training, with experiments in an environment with a parameter size of 12.8B showing up to a 58.8% decrease in memory usage compared to LoRA. To evaluate the generative performance of the fine-tuned model, texts generated from 100 inputs not included in the training data produced on average more than twice the number of words compared to the pre-trained model, with texts of positive sentiment also appearing more than twice as often. In a survey conducted for qualitative evaluation of generative performance, responses indicated that the fine-tuned model's generated outputs were more relevant to the given topics on average 77.5% of the time. This demonstrates that the positive review generation language model for sponsored content in this paper can enhance the efficiency of time management for content creation and ensure consistent marketing effects. However, to reduce the generation of content that deviates from the category of positive reviews due to elements of the pre-trained model, we plan to proceed with fine-tuning using the augmentation of training data.

• Korean Journal of Cognitive Science
• /
• v.4 no.2
• /
• pp.181-200
• /
• 1994

In the learning phase of multilyer feedforword neural network,there are problems such that local minimum,learning praralysis and slow learning speed when backpropagation algorithm used.To overcome these problems, the genetic algorithm has been used as learing method in the multilayer feedforword neural network instead of backpropagation algorithm.However,because the genetic algorith, does not have any mechanism for fine-tuned local search used in backpropagation method,it takes more time that the genetic algorithm converges to a global optimal solution.In this paper,we suggest a new GA-BP method which provides a fine-tunes local search to the genetic algorithm.GA-BP method uses gradient descent method as one of genetic algorithm's operators such as mutation or crossover.To show the effciency of the developed method,we applied it to the 3-parity bit problem with analysis.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.442-445
• /
• 1997

We present a hybrid self-tuning method of fuzzy inference systems with hyper elliptic Gaussian membership functions using genetic algorithm(GA) and back-propagation algorithm. The proposed self-tuning method has two phases : one is the coarse tuning process based on GA and the other is the fine tuning process based on back-propagation. But the parameters which is obtained by a GA are near optimal solutions. In order to solve the problem in GA applications, it uses a back-propagation algorithm, which is one of learning algorithms in neural networks, to finely tune the parameters obtained by a GA. We provide Box-Jenkins time series to evaluate the advantage and effectiveness of the proposed approach and compare with the conventional method.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.8
• /
• pp.1119-1126
• /
• 1995

The construction of rule-base for a nonlinear time-varying system, becomes much more complicated because of model uncertainty and parameter variations. Furthemore, FLC does not have an ability of adjusting rule- base in responding to some sudden changes of control environments. To cope with these problems, an auto-tuning method of the fuzzy rule-base is required. In this paper, the GA-based Fuzzy-Neural control system combining Fuzzy-Neural control theory with the genetic algorithm(GA), which is known to be very effective in the optimization problem, will be proposed. The tuning of the proposed system is performed by two tuning processes(the course tuning process and the fine tuning/adaptive learning process). The effectiveness of the proposed control system will be demonstrated by computer simulations using a two degree of freedom robot manipulator.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.289-294
• /
• 2023

대형 언어 모델(Large Language Model)은 방대한 데이터와 파라미터를 기반으로 문맥 이해에서 높은 성능을 달성하였지만, Human Alignment를 위한 문장 생성 제어 연구는 아직 활발한 도전 과제로 남아있다. 본 논문에서는 Instruction Tuning을 통한 문장 생성 제어 실험을 진행한다. 자연어 처리 도구를 사용하여 단일 혹은 다중 제약 조건을 포함하는 Instruction 데이터 셋을 자동으로 구축하고 한국어 언어 모델인 Polyglot-Ko 모델에 fine-tuning 하여 모델 생성이 제약 조건을 만족하는지 검증하였다. 실험 결과 4개의 제약 조건에 대해 평균 0.88의 accuracy를 보이며 효과적인 문장 생성 제어가 가능함을 확인하였다.

• Proceedings of the Korea Electromagnetic Engineering Society Conference
• /
• 2003.11a
• /
• pp.234-238
• /
• 2003

In this work, we have designed a fully integrated 2.4GHz LC-tuned voltage-controlled oscillator (VCO) with multiple tuning inputs for a $0.25-{\mu}m$ standard CMOS Process. The design of voltage-controlled oscillator is based on an LC-resonator with a spiral inductor of octagonal type and pMOS-varactors. Only two metal layer have been used in the designed inductor. The frequency tuning is achieved by using parallel pMOS transistors as varactors and back-gate tuned pMOS transistors in an active region. Coarse tuning is achieved by using 3-bit pMOS-varactors and fine tuning is performed by using back-gate tuned pMOS transistors in the active region. When 3-bit digital and analog inputs are applied to the designed circuits, voltage-controlled oscillator shows the tuning feature of frequency range between 2.3 GHz and 2.64 GHz. At the power supply voltage of 2.5 V, phase noise is -128dBc/Hz at 3MHz offset from the carrier, Total power dissipation is 7.5 mW.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.12
• /
• pp.4816-4834
• /
• 2020

This paper proposes transfer learning and fine-tuning techniques for a deep learning model to detect three distinct brain tumors from Magnetic Resonance Imaging (MRI) scans. In this work, the recent YOLOv4 model trained using a collection of 3064 T1-weighted Contrast-Enhanced (CE)-MRI scans that were pre-processed and labeled for the task. This work trained with the partial 29-layer YOLOv4-Tiny and fine-tuned to work optimally and run efficiently in most platforms with reliable performance. With the help of transfer learning, the model had initial leverage to train faster with pre-trained weights from the COCO dataset, generating a robust set of features required for brain tumor detection. The results yielded the highest mean average precision of 93.14%, a 90.34% precision, 88.58% recall, and 89.45% F1-Score outperforming other previous versions of the YOLO detection models and other studies that used bounding box detections for the same task like Faster R-CNN. As concluded, the YOLOv4-Tiny can work efficiently to detect brain tumors automatically at a rapid phase with the help of proper fine-tuning and transfer learning. This work contributes mainly to assist medical experts in the diagnostic process of brain tumors.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.20 no.4
• /
• pp.811-818
• /
• 2016

A continuous fine-tuning phase locked loop with an additional negative feedback loop has been proposed. When the phase locked loop is out-of-lock, the phase locked loop has a fast locking characteristic using the continuous band-selection loop. When the phase locked loop is near in-lock, the bandwidth is narrowed with the fine loop. The additional negative feedback loop consists of a voltage controlled oscillator, a frequency voltage converter and its internal loop filter. It serves a negative feedback function to the main phase locked loop, and improves the phase noise characteristics and the stability of the proposed phase locked loop. The additional negative feedback loop makes the continuous fine-tuning loop work stably without any voltage fluctuation in the loop filter. Measurement results of the fabricated phase locked loop in $0.18{\mu}m$ CMOS process show that the phase noise is -109.6dBc/Hz at 2MHz offset from 742.8MHz carrier frequency.

• Convergence Security Journal
• /
• v.23 no.5
• /
• pp.125-133
• /
• 2023

This paper proposes an AI model that determines parking and non-parking zones based on return authentication photos to address parking issues that may arise in shared electric scooter systems. In this study, we used a pre-trained Segformer_b0 model on ADE20K and fine-tuned it on tactile blocks and electric scooters to extract segmentation maps of objects related to parking and non-parking areas. We also presented a method to perform binary classification of parking and non-parking zones using the Swin model. Finally, after labeling a total of 1,689 images and fine-tuning the SegFomer model, it achieved an mAP of 81.26%, recognizing electric scooters and tactile blocks. The classification model, trained on a total of 2,817 images, achieved an accuracy of 92.11% and an F1-Score of 91.50% for classifying parking and non-parking areas.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.3
• /
• pp.570-590
• /
• 2024

Breast cancer ranks among the most prevalent forms of malignancy and foremost cause of death by cancer worldwide. It is not preventable. Early and precise detection is the only remedy for lowering the rate of mortality and improving the probability of survival for victims. In contrast to present procedures, thermography aids in the early diagnosis of cancer and thereby saves lives. But the accuracy experiences detrimental impact by low sensitivity for small and deep tumours and the subjectivity by physicians in interpreting the images. Employing deep learning approaches for cancer detection can enhance the efficacy. This study explored the utilization of thermography in early identification of breast cancer with the use of a publicly released dataset known as the DMR-IR dataset. For this purpose, we employed a novel approach that entails the utilization of a pre-trained MobileNetV2 model and fine tuning it through transfer learning techniques. We created three models using MobileNetV2: one was a baseline transfer learning model with weights trained from ImageNet dataset, the second was a fine-tuned model with an adaptive learning rate, and the third utilized early stopping with callbacks during fine-tuning. The results showed that the proposed methods achieved average accuracy rates of 85.15%, 95.19%, and 98.69%, respectively, with various performance indicators such as precision, sensitivity and specificity also being investigated.