• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.43-50
• /
• 2024

This paper evaluates the Korean sentiment analysis performance of large language models like GPT-3.5 and GPT-4 using a zero-shot approach facilitated by the ChatGPT API, comparing them to pre-trained Korean models such as KoBERT. Through experiments utilizing various Korean sentiment analysis datasets in fields like movies, gaming, and shopping, the efficiency of these models is validated. The results reveal that the LMKor-ELECTRA model displayed the highest performance based on F1-score, while GPT-4 particularly achieved high accuracy and F1-scores in movie and shopping datasets. This indicates that large language models can perform effectively in Korean sentiment analysis without prior training on specific datasets, suggesting their potential in zero-shot learning. However, relatively lower performance in some datasets highlights the limitations of the zero-shot based methodology. This study explores the feasibility of using large language models for Korean sentiment analysis, providing significant implications for future research in this area.

• Phonetics and Speech Sciences
• /
• v.15 no.3
• /
• pp.69-74
• /
• 2023

This study introduces an innovative model for zero-shot voice conversion that utilizes the capabilities of HuBERT. Zero-shot voice conversion models can transform the speech of one speaker to mimic that of another, even when the model has not been exposed to the target speaker's voice during the training phase. Comprising five main components (HuBERT, feature encoder, flow, speaker encoder, and vocoder), the model offers remarkable performance across a range of scenarios. Notably, it excels in the challenging unseen-to-unseen voice-conversion tasks. The effectiveness of the model was assessed based on the mean opinion scores and similarity scores, reflecting high voice quality and similarity to the target speakers. This model demonstrates considerable promise for a range of real-world applications demanding high-quality voice conversion. This study sets a precedent in the exploration of HuBERT-based models for voice conversion, and presents new directions for future research in this domain. Despite its complexities, the robust performance of this model underscores the viability of HuBERT in advancing voice conversion technology, making it a significant contributor to the field.

• Journal of IKEEE
• /
• v.27 no.4
• /
• pp.537-547
• /
• 2023

In many domains, lack of data inhibits adoption of advanced machine learning models. Recently, Few-Shot Learning (FSL) has been actively studied to tackle this problem. Utilizing prior knowledge obtained through observations on related domains, FSL achieved significant performance with only a few samples. In this paper, we present a survey on FSL in terms of data augmentation, embedding and metric learning, and meta-learning. In addition to interesting researches, we also introduce major benchmark datasets. FSL is widely adopted in various domains, but we focus on image analysis in this paper.

• Smart Structures and Systems
• /
• v.33 no.1
• /
• pp.55-64
• /
• 2024

This paper proposes a fusion imaging-based coating-defect classification method for steel structures that uses zero-shot learning. In the proposed method, a halogen lamp generates heat energy on the coating surface of a steel structure, and the resulting heat responses are measured by an infrared (IR) camera, while photos of the coating surface are captured by a charge-coupled device (CCD) camera. The measured heat responses and visual images are then analyzed using zero-shot learning to classify the coating defects, and the estimated coating defects are visualized throughout the inspection surface of the steel structure. In contrast to older approaches to coating-defect classification that relied on visual inspection and were limited to surface defects, and older artificial neural network (ANN)-based methods that required large amounts of data for training and validation, the proposed method accurately classifies both internal and external defects and can classify coating defects for unobserved classes that are not included in the training. Additionally, the proposed model easily learns about additional classifying conditions, making it simple to add classes for problems of interest and field application. Based on the results of validation via field testing, the defect-type classification performance is improved 22.7% of accuracy by fusing visual and thermal imaging compared to using only a visual dataset. Furthermore, the classification accuracy of the proposed method on a test dataset with only trained classes is validated to be 100%. With word-embedding vectors for the labels of untrained classes, the classification accuracy of the proposed method is 86.4%.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.722-725
• /
• 2023

대규모 언어 모델(LLM)은 대규모의 데이터를 학습하여 얻은 지식을 기반으로 텍스트와 다양한 콘텐츠를 인식하고 요약, 번역, 예측, 생성할 수 있는 딥러닝 알고리즘이다. 초기 공개된 LLM은 영어 기반 모델로 비영어권에서는 높은 성능을 기대할 수 없었으며, 이에 한국, 중국 등 자체적 LLM 연구개발이 활성화되고 있다. 본 논문에서는 언어가 LLM의 성능에 영향을 미치는가에 대하여 한국어 기반 LLM과 영어 기반 LLM으로 KoBEST의 4가지 Task에 대하여 성능비교를 하였다. 그 결과 한국어에 대한 사전 지식을 추가하는 것이 LLM의 성능에 영향을 미치는 것을 확인할 수 있었다.

• Journal of the Korea Convergence Society
• /
• v.12 no.11
• /
• pp.35-43
• /
• 2021

Recently, there has been a growing interest in zero-shot cross-lingual transfer, which leverages cross-lingual language models (CLLMs) to perform downstream tasks that are not trained in a specific language. In this paper, we point out the limitations of the data-centric aspect of quality estimation (QE), and perform zero-shot cross-lingual transfer even in environments where it is difficult to construct QE data. Few studies have dealt with zero-shots in QE, and after fine-tuning the English-German QE dataset, we perform zero-shot transfer leveraging CLLMs. We conduct comparative analysis between various CLLMs. We also perform zero-shot transfer on language pairs with different sized resources and analyze results based on the linguistic characteristics of each language. Experimental results showed the highest performance in multilingual BART and multillingual BERT, and we induced QE to be performed even when QE learning for a specific language pair was not performed at all.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.275-280
• /
• 2023

문서 그라운딩된 대화 시스템의 응답 성능 개선을 위한 방법론을 제안한다. 사전 학습된 거대 언어 모델 LLM(Large Language Model)인 Llama2 모델에 Zero-Shot In-Context learning을 적용하여 대화 마지막 유저 질문에 대한 응답을 생성하는 태스크를 수행하였다. 본 연구에서 제안한 응답 생성은 검색된 top-1 문서와 대화 기록을 참조해 초기 응답을 생성하고, 생성된 초기 응답을 기반으로 검색된 문서를 대상으로 재순위화를 수행한다. 이 후, 특정 순위의 상위 문서들을 이용해 최종 응답을 생성하는 과정으로 이루어진다. 검색된 상위 문서를 이용하는 응답 생성 방식을 Baseline으로 하여 본 연구에서 제안한 방식과 비교하였다. 그 결과, 본 연구에서 제안한 방식이 검색된 결과에 기반한 실험에서 Baseline 보다 F1, Bleu, Rouge, Meteor Score가 향상한 것을 확인 하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.439-441
• /
• 2023

전통적인 텍스트 분류 방법은 상당량의 라벨링된 데이터와 미리 정의된 클래스가 필요해서 그 적용성과 확장성이 제한된다. 그래서 이런 한계를 극복하기 위해 제로샷 러닝(Zero-shot Learning)이 등장했다. 텍스트 분류 분야에서 제로샷 텍스트 분류는 모델이 대상 클래스의 샘플을 미리 접하지 않고도 인스턴스를 분류할 수 있도록 하는 중요한 주제이다. 이 문제를 해결하기 위해 정책 네트워크를 활용한 심층 강화 학습(DRL) 기반 접근법을 제안한다. 이러한 방법을 통해 모델이 새로운 의미 공간에 효과적으로 적응하면서, 다른 모델들과 비교하여 제로샷 텍스트 분류의 정확도를 향상시킬 수 있었다. XLM-R 과 비교하면 최대 15.9%의 정확도 향상이 나타났다.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.11
• /
• pp.481-492
• /
• 2023

This study explores how to build a Korean dataset to extract information from text using generative large language models. In modern society, mixed information circulates rapidly, and effectively categorizing and extracting it is crucial to the decision-making process. However, there is still a lack of Korean datasets for training. To overcome this, this study attempts to extract information using text-based zero-shot learning using a generative large language model to build a purposeful Korean dataset. In this study, the language model is instructed to output the desired result through prompt engineering in the form of "system"-"instruction"-"source input"-"output format", and the dataset is built by utilizing the in-context learning characteristics of the language model through input sentences. We validate our approach by comparing the generated dataset with the existing benchmark dataset, and achieve 25.47% higher performance compared to the KLUE-RoBERTa-large model for the relation information extraction task. The results of this study are expected to contribute to AI research by showing the feasibility of extracting knowledge elements from Korean text. Furthermore, this methodology can be utilized for various fields and purposes, and has potential for building various Korean datasets.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.8
• /
• pp.195-204
• /
• 2023

In this study, prompt engineering that improves prompts was explored to improve the performance of teaching and learning materials recommendations using generative artificial intelligence such as GPT and Stable Diffusion. Picture materials were used as the types of teaching and learning materials. To explore the impact of the prompt composition, a Zero-Shot prompt, a prompt containing learning target grade information, a prompt containing learning goals, and a prompt containing both learning target grades and learning goals were designed to collect responses. The collected responses were embedded using Sentence Transformers, dimensionalized to t-SNE, and visualized, and then the relationship between prompts and responses was explored. In addition, each response was clustered using the k-means clustering algorithm, then the adjacent value of the widest cluster was selected as a representative value, imaged using Stable Diffusion, and evaluated by 30 elementary school teachers according to the criteria for evaluating teaching and learning materials. Thirty teachers judged that three of the four picture materials recommended were of educational value, and two of them could be used for actual classes. The prompt that recommended the most valuable picture material appeared as a prompt containing both the target grade and the learning goal.