• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.05a
• /
• pp.483-486
• /
• 2020

Age estimation and gender classification for human is a classic problem in computer vision. Almost research focus just only one task and the models are too heavy to run on low-cost system. In our research, we aim to apply multitasking learning to perform both task on a lightweight model which can achieve good precision on embedded system in the real time.

• ETRI Journal
• /
• v.43 no.6
• /
• pp.1004-1012
• /
• 2021

The collaboration productively interacting between multi-agents has become an emerging issue in real-world applications. In reinforcement learning, multi-agent environments present challenges beyond tractable issues in single-agent settings. This collaborative environment has the following highly complex attributes: sparse rewards for task completion, limited communications between each other, and only partial observations. In particular, adjustments in an agent's action policy result in a nonstationary environment from the other agent's perspective, which causes high variance in the learned policies and prevents the direct use of reinforcement learning approaches. Unexpected social loafing caused by high dispersion makes it difficult for all agents to succeed in collaborative tasks. Therefore, we address a paradox caused by the social loafing to significantly reduce total returns after a certain timestep of multi-agent reinforcement learning. We further demonstrate that the collaborative paradox in multi-agent environments can be avoided by our proposed effective early stop method leveraging a metric for social loafing.

• Journal of Broadcast Engineering
• /
• v.23 no.3
• /
• pp.351-360
• /
• 2018

Human emotion recognition is a research topic that is receiving continuous attention in computer vision and artificial intelligence domains. This paper proposes a method for classifying human emotions through multiple neural networks based on multi-modal signals which consist of image, landmark, and audio in a wild environment. The proposed method has the following features. First, the learning performance of the image-based network is greatly improved by employing both multi-task learning and semi-supervised learning using the spatio-temporal characteristic of videos. Second, a model for converting 1-dimensional (1D) landmark information of face into two-dimensional (2D) images, is newly proposed, and a CNN-LSTM network based on the model is proposed for better emotion recognition. Third, based on an observation that audio signals are often very effective for specific emotions, we propose an audio deep learning mechanism robust to the specific emotions. Finally, so-called emotion adaptive fusion is applied to enable synergy of multiple networks. The proposed network improves emotion classification performance by appropriately integrating existing supervised learning and semi-supervised learning networks. In the fifth attempt on the given test set in the EmotiW2017 challenge, the proposed method achieved a classification accuracy of 57.12%.

• The Research Journal of the Costume Culture
• /
• v.30 no.5
• /
• pp.656-674
• /
• 2022

The purpose of this study is to reveal and compare the differences in the types and characteristics of purchase channel journeys of MZ generation consumers. In this study a survey was conducted on the purchase channel journey of 20 women in the MZ generation using the ethnographic method of in-depth interviews and observations. As a result, three purchase channel journeys were identified: mobile, multi-channel, and offline. These were variously subdivided according to the characteristics of the MZ generations. Gen Z's journey was categorized into types: fashion platform app, Youtube, multi-channel supplement, multi-channel non-planned store visit, offline loyalty store, and impulsive offline store. Gen M's journey was categorized as: an online community bond, portal site, online loyalty store, multi-channel brand involvement, multi-channel efficiency, a multi-channel conversion, offline efficiency and offline task. The difference in mobile journey between generations was found in the time and length of the purchase. Gen M recognized both online and offline search processes to be tiring, while Gen Z enjoyed the search process using the online path. In the offline journey Gen Z began with their own intention to purchase, while Gen M sometimes recognized that purchasing fashion products necessary for work was a cumbersome task.

• Journal of Broadcast Engineering
• /
• v.24 no.6
• /
• pp.1122-1133
• /
• 2019

We have designed and implemented a new data processing framework called MOHA(Mtc On HAdoop) which can effectively support Many-Task Computing(MTC) applications in a YARN-based Hadoop platform. MTC applications can be composed of a very large number of computational tasks ranging from hundreds of thousands to millions of tasks, and each MTC application may have different resource usage patterns. Therefore, we have implemented MOHA-TaskExecutor(a pilot-job that executes real MTC application tasks)'s Adaptive Parallel Computability which can adaptively execute multiple tasks simultaneously, in order to improve the parallel computability of a YARN container and the overall system throughput. We have implemented multi-threaded version of TaskExecutor which can "independently and dynamically" adjust the number of concurrently running tasks, and in order to find the optimal number of concurrent tasks, we have employed Hill-Climbing algorithm.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.58-61
• /
• 2002

In this paper, we present a new generalized k-Tree-based (CU/PE) allocation model to perform dynamic resource (CU/PE) allocation/deallocation decision for the reconfigurable MSIMD/MIMD multi-dimensional (k-D) mesh-connected architectures. Those reconfigurable multi-SIMD/MIMD systems allow dynamic modes of executing tasks, which are SIMD and MIMD. The MIMD task requires only the free sub-system; however the SIMD task needs not only the free sub-system but also the corresponding free CU. In our new k-Tree-based (CU/PE) allocation model, we introduce two best-fit heuristics for the CU allocation decision: 1) the CU depth first search (CU-DFS) in O(kN$_{f}$ ) time and 2) the CU adjacent search (CU-AS) in O(k2$^{k}$ ) time. By the simulation study, the system performance of these two CU allocation strategies was also investigated. Our simulation results showed that the CU-AS and CU-DFS strategies performed the same system performance when applied for the reconfigurable MSIMD/MIMD 2-D and 3-D mesh-connected architectures.

• Structural Monitoring and Maintenance
• /
• v.2 no.3
• /
• pp.181-197
• /
• 2015

Assessing the condition of paint on civil structures is an important but challenging and costly task, in particular when it comes to large and complex structures. Current practices of visual inspection are labour-intensive and time-consuming to perform. In addition, this task usually relies on the experience and subjective judgment of individual inspectors. In this study, hyperspectral imaging and classification techniques are proposed as a method to objectively assess the state of the paint on a civil or other structure. The ultimate objective of the work is to develop a technology that can provide precise and automatic grading of paint condition and assessment of degradation due to age or environmental factors. Towards this goal, we acquired hyperspectral images of steel surfaces located at long (mid-range) and short distances on the Sydney Harbour Bridge with an Acousto-Optics Tunable filter (AOTF) hyperspectral camera (consisting of 21 bands in the visible spectrum). We trained a multi-class Support Vector Machines (SVM) classifier to automatically assess the grading of the paint from hyperspectral signatures. Our results demonstrate that the classifier generates highly accurate assessment of the paint condition in comparison to the judgement of human experts.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.6
• /
• pp.670-677
• /
• 2019

As the presence of background noise in acoustic signal degrades the performance of speech or acoustic event recognition, it is still challenging to extract noise-robust acoustic features from noisy signal. In this paper, we propose a combined structure of Wasserstein Generative Adversarial Network (WGAN) and MultiTask AutoEncoder (MTAE) as deep learning architecture that integrates the strength of MTAE and WGAN respectively such that it estimates not only noise but also speech features from noisy acoustic source. The proposed MTAE-WGAN structure is used to estimate speech signal and the residual noise by employing a gradient penalty and a weight initialization method for Leaky Rectified Linear Unit (LReLU) and Parametric ReLU (PReLU). The proposed MTAE-WGAN structure with the adopted gradient penalty loss function enhances the speech features and subsequently achieve substantial Phoneme Error Rate (PER) improvements over the stand-alone Deep Denoising Autoencoder (DDAE), MTAE, Redundant Convolutional Encoder-Decoder (R-CED) and Recurrent MTAE (RMTAE) models for robust speech recognition.

• Korean Journal of Social Welfare
• /
• v.62 no.4
• /
• pp.59-77
• /
• 2010

The purpose of this paper is to suggest the implications of volunteering's task design and the volunteer's placement that make the motive-benefit matching from the functional perspective of motivation. For this study, we conducted a multi-regression analysis to examine the impact of the volunteer motivation on task satisfaction and retention will. Results showed that first, volunteers' most important motive for volunteering was social motive, followed by enhancement, value. Secondly, each motive influenced the task satisfaction and retention will of volunteers differentially. The practical implications of these findings were discussed.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.12
• /
• pp.4372-4388
• /
• 2014

Robotic sensor network (RSN) contains mobile sensors and robots providing feasible solution for many multi-agent applications. One of the most critical issues in RSN and its application is how to effectively assign tasks. This paper presents a novel connectivity preserving hybrid task allocation strategy to answer the question particularly for RSN. Firstly, we model the task allocation in RSN to distinguish the discovering and allocating processes. Secondly, a fully distributed simple Task-oriented Unoccupied Neighbor Algorithm, named TUNA, is developed to allocate tasks with only partial view of the network topology. A connectivity controller is finally developed and integrated into the strategy to guarantee the global connectivity of entire RSN, which is critical to most RSN applications. The correctness, efficiency and scalability of TUNA are proved with both theoretical analysis and experimental simulations. The evaluation results show that TUNA can effectively assign tasks to mobile robots with the requirements of only a few messages and small movements of mobile agents.