• Journal of Intelligence and Information Systems
• /
• v.28 no.3
• /
• pp.45-55
• /
• 2022

Recently, large-scale language models (LPLM) have been shown state-of-the-art performances in various tasks of natural language processing including intent classification. However, fine-tuning LPLM requires much computational cost for training and inference which is not appropriate for dialog system. In this paper, we propose compressed intent classification model for multi-agent in low-resource like CPU. Our method consists of two stages. First, we trained sentence encoder from LPLM then compressed it through knowledge distillation. Second, we trained agent-specific adapter for intent classification. The results of three intent classification datasets show that our method achieved 98% of the accuracy of LPLM with only 21% size of it.

• Culinary science and hospitality research
• /
• v.23 no.6
• /
• pp.78-88
• /
• 2017

The main goal of this study was to demonstrate the causal relationships between multi-dimensional work-values, job satisfaction and task performance of Y generational employee in the Korean hotel industry. In order to achieve the purpose, the data were collected from Y generational employees working in the major cities in Korea. To analyze the data, frequency, exploratory factor analysis, reliability analysis, PROCESS Macro analysis and simple regression analysis were undertaken by using SPSS 18.0. The results were as follows; 1. Work-values were found out as five factors; Prestige workvalues, Personal development work-values, Work condition work-values, Personal welfare work-values, Social/ Altruistic work-values. 2. All five factors were found to be significant in enhancing job satisfaction. 3. Two factors which were prestige work-values and personal welfare work-values were found to be significant in facilitating task performance. 4. Job satisfaction was revealed as a mediator between all work-value factors and task performance. This study provided practical information about work-values of Y generational hoteliers to positively affect their job satisfaction and task performance. This study also confirmed that it is important to enhance job satisfaction in order to make employees perform their service duties better.

• Korean Journal of Applied Biomechanics
• /
• v.30 no.3
• /
• pp.265-273
• /
• 2020

Objective: The aim of this study was to investigate the effect of participation in contact sports on neurocognitive scores, dual-task walking velocity, and cognitive costs in retired athletes. Method: Forty-four retired athletes (mean age = 26.4±5.5 yrs) and thirty-eight controls (mean age = 26.1±4.9 yrs) participated in this study. Neurocognitive score was collected using computerized neurocognitive testing using RehaCom. Gait velocity was collected one single task, four dual-tasks, and two multi-tasks using Optogait. Mann-Whitney U test was performed to compared differences in cognitive scores among groups. A mixed-design two-way ANOVA and Bonferroni posthoc test were used to assess the effect of group and walking tasks for each condition. Results: The auditory divided attention of neurocognitive score of retired athletes was higher than the control group (p < 0.05). No statistical differences were observed in the other neurocognitive scores between groups. The changes in walking velocity and cognitive costs according to the dual-task walking tests differed between the two groups (p < 0.05). Conclusion: Although participation in contact sports did not affect the neurocognitive results of retired athletes, it could be confirmed that the reduction in walking velocity and an increase in cognitive costs during dual-task walking. Rather than observing only neurocognitive scores as a single evaluation item for cognitive evaluation of retired athletes in relation to daily life, the application of the dual-task gait test may provide useful information.

• Journal of KIISE
• /
• v.44 no.3
• /
• pp.223-231
• /
• 2017

In this paper, we study the performance of recently popular multi-core systems in mobiles. Previous research on the multi-core performance usually focused on the desktop PC. However, there is enough scope to further analyze heterogeneous multi-core systems. Therefore, by extending homogeneous multi-core systems, we analyze the heterogeneous multi-core systems using Size Interval Task Allocation (SITA) for job allocation, and Shortest Remaining Processing Time (SRPT) scheduling, for each individual core. We propose a new computational method regarding the cutoff point, which is crucial in analyzing SITA, by calculating the sojourn time. This facilitate easy and accurate calculation of the sojourn time. We further confirm our analysis through the ESESC simulator that provides actual measurements.

• Smart Structures and Systems
• /
• v.8 no.1
• /
• pp.107-117
• /
• 2011

Biological sensory systems continuously monitor and analyze changes in real-world environments that are relevant to an animal's specific behavioral needs and goals. Understanding the sensory mechanisms and information processing principles that biological systems utilize for efficient sensory data acquisition may provide useful guidance for the design of smart-sensing systems in engineering applications. Weakly electric fish, which use self-generated electrical energy to actively sense their environment, provide an excellent model system for studying biological principles of sensory data acquisition. The electrosensory system enables these fish to hunt and navigate at night without the use of visual cues. To achieve reliable, real-time task performance, the electrosensory system implements a number of smart sensing strategies, including efficient stimulus encoding, multi-scale virtual sensor arrays, task-dependent filtering and online subtraction of sensory expectation.

• Management Science and Financial Engineering
• /
• v.19 no.2
• /
• pp.1-11
• /
• 2013

An interesting puzzle in business practices is that although many researchers emphasize the benefits of a revenue-sharing contract, a wholesale contract has remained to be the most common contractual form. By introducing the concept of unobservable efforts, we examine the contractual efficiency of a wholesale contract and a revenue-sharing contract. The multi-task agency model and experimental design approach are used to analyze the relationship between the contractual efficiency and parameters. A major finding of our study is that a wholesale contract coordinates unobservable efforts, while it fails to coordinate the order quantity decision. Because unobservable efforts have mixed effects on the contractual efficiency, the superiority of contract type depends on parameters. This finding implies that a wholesale contract can be a competitive contract, especially when unobservable efforts are heavily involved. Our conclusion is that the current popularity of a wholesale contract is manager's rational response to complex supply chain environments rather than irrational behaviors.

• Journal of Institute of Control, Robotics and Systems
• /
• v.18 no.3
• /
• pp.218-223
• /
• 2012

Maintenance and improvement of the graph connectivity is very important for decentralized multi-agent systems. Although the CBBA (Consensus-Based Bundle Algorithm) guarantees suboptimal performance and bounded convergence time, it is only valid for connected graphs. In this study, we apply a decentralized estimation procedure that allows each agent to track the algebraic connectivity of a time-varying graph. Based on this estimation, we design a decentralized gradient controller to maintain the graph connectivity while agents are traveling to perform assigned tasks. Simulation result for fully-actuated first-order agents that move in a 2-D plane are presented.

• Journal of the Korean Society of Safety
• /
• v.24 no.1
• /
• pp.69-77
• /
• 2009

This paper reports the aging driver traffic accident severity modeling results. For the modeling, Poisson regression approach is applied using the data set obtained from the Korea Transportation Safety Authority's simulator-based driver aptitude test results. The test items include the estimations of moving objects' speed and stopping distance, drivers' multi-task capability, and kinetic depth perception and so on. The resulting model with the response variable of equivalent property damage only(EPDO) indicated that EPDO is significantly influenced by moving objects' speed estimation and drivers' multi-task capabilities. More interestingly, a comparison with the younger driver model revealed that the degradation of such capabilities may result in severer crashes for older drivers as suggested by the higher estimated parameters for the older driver model.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.45-47
• /
• 2022

Mobile edge computing (MEC), which enables mobile terminals to offload computational tasks to a server located at the user's edge, is considered an effective way to reduce the heavy computational burden and achieve efficient computational offloading. In this paper, we study a multi-user MEC system in which multiple user devices (UEs) can offload computation to the MEC server via a wireless channel. To solve the resource allocation and task offloading problem, we take the total cost of latency and energy consumption of all UEs as our optimization objective. To minimize the total cost of the considered MEC system, we propose an DRL-based method to solve the resource allocation problem in wireless MEC. Specifically, we propose a Asynchronous Advantage Actor-Critic (A3C)-based scheme. Asynchronous Advantage Actor-Critic (A3C) is applied to this framework and compared with DQN, and Double Q-Learning simulation results show that this scheme significantly reduces the total cost compared to other resource allocation schemes

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.05a
• /
• pp.437-440
• /
• 2021

지난 10년 동안 기계학습을 통해 자연어 처리 분야에서 많은 발전이 있었다. Machine translation, question answering과 같은 문제는 사용 가능한 데이터가 많은 언어에서 높은 정확도 성능 결과를 보여준다. 그러나 low-resource 언어에선 동일한 수준의 성능에 도달할 수 없다. 카자흐어는 형태학적 분석을 위해 구축된 대용량 데이터셋이 없으므로 low-resource 환경이다. 카자흐어는 단일 어근으로 수백 개의 단어 형태를 생성할 수 있는 교착어이다. 그래서 카자흐어 문장의 형태학적 분석은 카자흐어 문장의 의미를 이해하는 기본적인 단계이다. 기존에 존재하는 카자흐어 데이터셋은 구체적인 형태학적 분석의 부재로 모델이 충분한 학습이 이루어지지 못하기 때문에 본 논문에서 새로운 데이터셋을 제안한다. 본 논문은 low-resource 환경에서 높은 정확도를 달성할 수 있는 신경망 모델 기반의 카자흐어 형태학 분석기를 제안한다.