• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.251-273
• /
• 2022

Recently, research on applying text analysis to deep learning has steadily continued. In particular, researches have been actively conducted to understand the meaning of words and perform tasks such as summarization and sentiment classification through a pre-trained language model that learns large datasets. However, existing pre-trained language models show limitations in that they do not understand specific domains well. Therefore, in recent years, the flow of research has shifted toward creating a language model specialized for a particular domain. Domain-specific pre-trained language models allow the model to understand the knowledge of a particular domain better and reveal performance improvements on various tasks in the field. However, domain-specific further pre-training is expensive to acquire corpus data of the target domain. Furthermore, many cases have reported that performance improvement after further pre-training is insignificant in some domains. As such, it is difficult to decide to develop a domain-specific pre-trained language model, while it is not clear whether the performance will be improved dramatically. In this paper, we present a way to proactively check the expected performance improvement by further pre-training in a domain before actually performing further pre-training. Specifically, after selecting three domains, we measured the increase in classification accuracy through further pre-training in each domain. We also developed and presented new indicators to estimate the specificity of the domain based on the normalized frequency of the keywords used in each domain. Finally, we conducted classification using a pre-trained language model and a domain-specific pre-trained language model of three domains. As a result, we confirmed that the higher the domain specificity index, the higher the performance improvement through further pre-training.

• Journal of KIISE:Computing Practices and Letters
• /
• v.5 no.4
• /
• pp.467-478
• /
• 1999

실시간 태스크의 최악 실행시간을 예측할 때 과예측이 발생하는 원인은, 첫째 프로그램의 동적인 최악 실행 행태를 정적으로 분석하는 것이 근본적으로 어렵기 때문이며, 둘째 최근의 RISC 형태 프로세서에 포함되어 있는 파이프라인 실행 구조와 캐쉬 등이 그러한 정적 분석을 더욱 어렵게 만들기 때문이다. 그런데 기존의 연구에서는 각각의 과예측 원인을 해결하기 위한 방법에 대해서만 언급하고 있을 뿐 분석의 정확도에서 각 원인이 차지하는 비중에 대해서는 언급하고 있지 않다. 이에 본 연구에서는 최악 실행시간 예측시 과예측을 유발하는 원인들, 즉 분석 요소들의 영향을 정량적으로 조사함으로써 기존의 최악 실행시간 분석 기법들이 보완해야 할 방향을 제시하고자 한다. 본 연구에서는 실험이 특정 분석 기법에 의존하지 않도록 하기 위하여 시뮬레이션 방법에 기반한다. 이를 위해 분석 요소별 스위치가 포함된 MIPS R3000 프로세서를 위한 시뮬레이터를 구현하였는데, 각 스위치는 해당 분석 요소에 대한 분석의 정확도 수준을 결정한다. 모든 스위치 조합에 대해서 시뮬레이션을 반복 수행한 다음 분산 분석을 수행하여 어떤 분석 요소가 가장 큰 영향을 끼치는지 고찰한다.Abstract Existing analysis techniques for estimating the worst case execution time (WCET) of real-time tasks still suffer from significant overestimation due to two types of overestimation sources. First, it is unavoidably difficult to predict dynamic behavior of programs statically. Second, pipelined execution and caching found in recent RISC-style processors even more complicate such a prediction. Although these overestimation sources have been attacked in many existing analysis techniques, we cannot find in the literature any description about questions like which one is most important. Thus, in this paper, we quantitatively analyze the impacts of overestimation sources on the accuracy of the worst case timing analysis. Using the results, we can identify dominant overestimation sources that should be analyzed more accurately to get tighter WCET estimations. To make our method independent of any existing analysis techniques, we use simulation based methodology. We have implemented a MIPS R3000 simulator equipped with several switches, each of which determines the accuracy level of the timing analysis for the corresponding overestimation source. After repeating simulation for all of the switch combinations, we perform the variance analysis and study which factor has the largest impact on the accuracy of the predicted WCETs.