• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.11
• /
• pp.5546-5559
• /
• 2019

Facial action units (AUs) have recently drawn increased attention because they can be used to recognize facial expressions. A variety of methods have been designed for frontal-view AU detection, but few have been able to handle multi-view face images. In this paper we propose a method for multi-view facial AU detection using a fused multilayer, multi-task, and multi-label deep learning network. The network can complete two tasks: AU detection and facial view detection. AU detection is a multi-label problem and facial view detection is a single-label problem. A residual network and multilayer fusion are applied to obtain more representative features. Our method is effective and performs well. The F1 score on FERA 2017 is 13.1% higher than the baseline. The facial view recognition accuracy is 0.991. This shows that our multi-task, multi-label model could achieve good performance on the two tasks.

• Journal of Information Technology Services
• /
• v.16 no.3
• /
• pp.103-111
• /
• 2017

The real time scheduling is a key research area in high performance computing and has been a source of challenging problems. A periodic task is an infinite sequence of task instance where each job of a task comes in a regular period. The RMS (Rate Monotonic Scheduling) algorithm has the advantage of a strong theoretical foundation and holds out the promise of reducing the need for exhaustive testing of the scheduling. Many real-time systems built in the past based their scheduling on the Cyclic Executive Model because it produces predictable schedules which facilitate exhaustive testing. In this work we propose hybrid scheduling method which combines features of both of these scheduling algorithms. The original rate monotonic scheduling algorithm didn't consider the uniform sampling tasks in the real time systems. We have enumerated some issues when the RMS is applied to our hybrid scheduling method. We found the scheduling bound for the hard real-time systems which include the uniform sampling tasks. The suggested hybrid scheduling algorithm turns out to have some advantages from the point of view of the real time system designer, and is particularly useful in the context of large critical systems. Our algorithm can be useful for real time system designer who must guarantee the hard real time tasks.

• Journal of the Korea Society of Computer and Information
• /
• v.13 no.3
• /
• pp.153-160
• /
• 2008

One of the most important features of the Mobile Manipulator is redundant freedom. Using it's redundant freedom, a Mobile Manipulator can move in various modes, and perform dexterous motions. In this paper, to improve robot job performance, two robots -mobile robot, task robot- are joined together to perform a job, we studied the optimal position and posture of a Mobile Manipulator to achieve a minimum of movement of each robot joint. Kinematics of mobile robot and task robot is solved. Using the mobility of a Mobile robot, the weight vector of robots is determined. Using the Gradient method, global motion trajectory is minimized, so the job which the Mobile Manipulator performs is optimized. The proposed algorithm is verified with PURL-II which is Mobile Manipulator combined Mobile robot and task robot, and the results are discussed.

• The Journal of Korean Academic Society of Nursing Education
• /
• v.20 no.4
• /
• pp.482-490
• /
• 2014

Purpose: This research compares self-leadership, team efficacy, problem solving processes and task satisfaction in response to teaching methods applied to nursing students, and determines whether variations exist. Method: This research experiments before and after the training of a nonequivalent group. The subjects were 36 learners of action learning methods and 39 learners of nursing course methods, and the research took place from October through December 2012. Results: Prior to the training, the general features and measurable variables of the two groups of subjects were similar, and self-leadership, team efficacy, problem solving process and task satisfaction in both groups were elevated compared to pre-training. In particular, in comparison with the nursing course, there was a notable difference in scores, the action learning method receiving high scores in the problem solving process (t=2.92, p=.005) and task satisfaction (t=2.54, p=.013) Conclusion: It is recommended that educators not only conduct the practice training course for teaching methods, but also incorporate action learning.

• Journal of Institute of Control, Robotics and Systems
• /
• v.15 no.5
• /
• pp.538-542
• /
• 2009

Industrial manipulators have been mostly used in large companies such as automakers and electronics companies. In recent years, however, demands for industrial manipulators from small and medium-sized enterprises are on the increase because of shortage of manpower and high wages. Since these companies cannot hire robot engineers for operation and programming of a robot, intuitive teaching and playback techniques of a robot manipulator should replace the robot programming which requires substantial knowledge of a robot. This paper proposes an intuitive teaching and playback algorithm used in assembly tasks. An operator can directly teach the robot by grasping the end-effector and moving it to the desired point in the teaching phase. The 6 axis force/torque sensor attached to the manipulator end-effector is used to sense the human intention in teaching the robot. After this teaching phase, a robot can track the target position or trajectory accurately in the playback phase. When the robot contacts the environment during the teaching and playback phases, impedance control is conducted to make the contact task stable. Peg-in-hole experiments are selected to validate the proposed algorithm since this task can describe the important features of various assembly tasks which require both accurate position and force control. It is shown that the proposed teaching and playback algorithm provides high positioning accuracy and stable contact tasks.

• International Journal of Advanced Culture Technology
• /
• v.5 no.1
• /
• pp.83-89
• /
• 2017

In this paper, we propose an extended method of one-hot Long Short-Term Memory (LSTM) and evaluate the performance on spam filtering task. Most of traditional methods proposed for spam filtering task use word occurrences to represent spam or non-spam messages and all syntactic and semantic information are ignored. Major issue appears when both spam and non-spam messages share many common words and noise words. Therefore, it becomes challenging to the system to filter correct labels between spam and non-spam. Unlike previous studies on information filtering task, instead of using only word occurrence and word context as in probabilistic models, we apply a neural network-based approach to train the system filter for a better performance. In addition to one-hot representation, using term weight with attention mechanism allows classifier to focus on potential words which most likely appear in spam and non-spam collection. As a result, we obtained some improvement over the performances of the previous methods. We find out using region embedding and pooling features on the top of LSTM along with attention mechanism allows system to explore a better document representation for filtering task in general.

• Korean Journal of Cognitive Science
• /
• v.23 no.2
• /
• pp.225-248
• /
• 2012

The summed-similarity model of visual short-term recognition proposes that the estimated amount of summed similarity between remembered items and a recognition probe determines recognition judgement decision (Kahan & Sekuler, 2002). This study examined the effect of a task-irrelevant location change on the recognition decision against two remembered Gabor gratings differing in their spatial frequencies. On each trial in Experiment, participants reported if two gratings displayed across the visual fields are the same or not as the probe grating displayed after about a second of memory delay. The probe grating would be the same as or different from the memory items (lure) by 1 or 4 JND units. The location of the probe would also vary randomly across the left and right visual field with respect to the location of the corresponding memory item. The participants were instructed to perform their recognition task exclusively to the spatial frequencies of the memory items and the probe while ignoring the potential location change of the probe. The results showed that false-recognition rates of the lure probe increased as the summed similarity between the memory items and the probe increased. The rates also further increased in the condition where the probe location was different from the location of the corresponding memory item compared to the condition where the probe location was the same. The increased false-recognition rates indicate that information stored into visual short-term memory is represented as a form of well-bound visual features rather than independent features.

• ETRI Journal
• /
• v.42 no.1
• /
• pp.90-100
• /
• 2020

Named entity recognition (NER) continues to be an important task in natural language processing because it is featured as a subtask and/or subproblem in information extraction and machine translation. In Urdu language processing, it is a very difficult task. This paper proposes various deep recurrent neural network (DRNN) learning models with word embedding. Experimental results demonstrate that they improve upon current state-of-the-art NER approaches for Urdu. The DRRN models evaluated include forward and bidirectional extensions of the long short-term memory and back propagation through time approaches. The proposed models consider both language-dependent features, such as part-of-speech tags, and language-independent features, such as the "context windows" of words. The effectiveness of the DRNN models with word embedding for NER in Urdu is demonstrated using three datasets. The results reveal that the proposed approach significantly outperforms previous conditional random field and artificial neural network approaches. The best f-measure values achieved on the three benchmark datasets using the proposed deep learning approaches are 81.1%, 79.94%, and 63.21%, respectively.

• Journal of the Korean Society of Safety
• /
• v.32 no.5
• /
• pp.179-185
• /
• 2017

Accident analysis is an essential process to make basic data for accident prevention. Most researches depend on survey data and accident statistics to analyze accidents, but these kinds of data are not sufficient for systematic and detailed analysis. We, in this paper, propose an accident classification model that extracts task type, original cause materials, accident type, and the number of deaths from accident reports. The classification model is a support vector machine (SVM) with word occurrence features, and these features are selected based on mutual information. Experiment shows that the proposed model can extract task type, original cause materials, accident type, and the number of deaths with almost 100% accuracy. We also develop an accident ontology to express the information extracted by the classification model. Finally, we illustrate how the proposed classification model and ontology effectively works for the accident analysis. The classification model and ontology are expected to effectively analyze various accidents.

• Journal of KIISE:Software and Applications
• /
• v.31 no.10
• /
• pp.1412-1417
• /
• 2004

Sparse coding or independent component analysis (ICA) which is a holistic representation, was successfully applied to elucidate early auditor${\gamma}$ processing and to the task of sound classification. In contrast, parts-based representation is an alternative way o) understanding object recognition in brain. In this thesis we employ the non-negative matrix factorization (NMF) which learns parts-based representation in the task of sound classification. Methods of feature extraction from the spectro-temporal sounds using the NMF in the absence or presence of noise, are explained. Experimental results show that NMF-based features improve the performance of sound classification over ICA-based features.