• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.9
• /
• pp.4665-4683
• /
• 2019

Previous methods build image annotation model by leveraging three basic dependencies: relations between image and label (image/label), between images (image/image) and between labels (label/label). Even though plenty of researches show that multiple dependencies can work jointly to improve annotation performance, different dependencies actually do not "work jointly" in their diagram, whose performance is largely depending on the result predicted by image/label section. To address this problem, we propose the adaptive attention annotation model (AAAM) to associate these dependencies with the prediction path, which is composed of a series of labels (tags) in the order they are detected. In particular, we optimize the prediction path by detecting the relevant labels from the easy-to-detect to the hard-to-detect, which are found using Binary Cross-Entropy (BCE) and Triplet Margin (TM) losses, respectively. Besides, in order to capture the inforamtion of each label, instead of explicitly extracting regional featutres, we propose the self-attention machanism to implicitly enhance the relevant region and restrain those irrelevant. To validate the effective of the model, we conduct experiments on three well-known public datasets, COCO 2014, IAPR TC-12 and NUSWIDE, and achieve better performance than the state-of-the-art methods.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.34C no.9
• /
• pp.43-52
• /
• 1997

Processors using the superscalar rchitecture can achieve high performance by executing multipel instructions in a clock cycle. It is made possible by having multiple functional units and issuing multiple instructions to functional units simultaneously. But instructions can be dependent on one another and these dependencies prevent some instructions form being issued at the same cycle. In this paper, we designed an issue unit of a superscalar RISC microprocessor that can issue four instructions per cycle. The issue unit receives instructions form a prefetch unit, and issues them in order at a rate of as high as four instructions in one cycle for maximum utilization of functional units. By using an instruction buffer, the unit decouples instruction fetch and issue to improve instruction ussue rate. The issue unit is composed of an instruction buffer and an instruction decoder. The instruction buffer aligns and stores instructions from the prefetch unit, and sends the earliest four available isstructions to the instruction decoder. The instruction decoder decodes instructions, and issues them if they are free form data dependencies and necessary functional units and rgister file prots are available. The issue unit is described with behavioral level HDL (lhardware description language). The result of simulation using C programs shows that instruction issue rate is improved as the instruction buffer size increases, and 12-entry instruction buffer is found to be optimum considering performance and hardware cost of the instruction buffer.

• International journal of advanced smart convergence
• /
• v.12 no.4
• /
• pp.88-97
• /
• 2023

Traffic flow prediction is of great significance in urban planning and traffic management. As the complexity of urban traffic increases, existing prediction methods still face challenges, especially for the fusion of spatiotemporal information and the capture of long-term dependencies. This study aims to use the fusion model of graph neural network to solve the spatio-temporal information fusion problem in traffic flow prediction. We propose a new deep learning model Spatio-Temporal Information Fusion using Graph Neural Networks (STFGNN). We use GCN module, TCN module and LSTM module alternately to carry out spatiotemporal information fusion. GCN and multi-core TCN capture the temporal and spatial dependencies of traffic flow respectively, and LSTM connects multiple fusion modules to carry out spatiotemporal information fusion. In the experimental evaluation of real traffic flow data, STFGNN showed better performance than other models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.6 no.1
• /
• pp.229-240
• /
• 2012

Advances in MEMS and CMOS technologies have motivated the development of low cost/power sensors and wireless multimedia sensor networks (WMSN). The WMSNs were created to ubiquitously harvest multimedia content. Such networks have allowed researchers and engineers to glimpse at new Machine-to-Machine (M2M) Systems, such as remote monitoring of biosignals for telemedicine networks. These systems require the acquisition of a large number of data streams that are simultaneously generated by multiple distributed devices. This paradigm of data generation and transmission is known as event-streaming. In order to be useful to the application, the collected data requires a preprocessing called data fusion, which entails the temporal alignment task of multimedia data. A practical way to perform this task is in a centralized manner, assuming that the network nodes only function as collector entities. However, by following this scheme, a considerable amount of redundant information is transmitted to the central entity. To decrease such redundancy, data fusion must be performed in a collaborative way. In this paper, we propose a collaborative data alignment approach for event-streaming. Our approach identifies temporal relationships by translating temporal dependencies based on a timeline to causal dependencies of the media involved.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.1
• /
• pp.1-14
• /
• 2024

Time-series forecasting is extensively used in the actual world. Recent research has shown that Transformers with a self-attention mechanism at their core exhibit better performance when dealing with such problems. However, most of the existing Transformer models used for time series prediction use the traditional encoder-decoder architecture, which is complex and leads to low model processing efficiency, thus limiting the ability to mine deep time dependencies by increasing model depth. Secondly, the secondary computational complexity of the self-attention mechanism also increases computational overhead and reduces processing efficiency. To address these issues, the paper designs an efficient multi-layer attention-based time-series forecasting model. This model has the following characteristics: (i) It abandons the traditional encoder-decoder based Transformer architecture and constructs a time series prediction model based on multi-layer attention mechanism, improving the model's ability to mine deep time dependencies. (ii) A cross attention module based on cross attention mechanism was designed to enhance information exchange between historical and predictive sequences. (iii) Applying a recently proposed sparse attention mechanism to our model reduces computational overhead and improves processing efficiency. Experiments on multiple datasets have shown that our model can significantly increase the performance of current advanced Transformer methods in time series forecasting, including LogTrans, Reformer, and Informer.

• Journal of Korean Society for Quality Management
• /
• v.23 no.3
• /
• pp.84-92
• /
• 1995

A method is proposed for the simultaneous optimization of several response functions that depend on the same set of controllable variables and are adequately represented by a response surface model (polynomial regression model) with the same degree and with constraint that the individual responses have the target values. First, the multiple responses data are checked for linear dependencies among the responses by eigenvalue analysis. Thus a set of responses with no linear functional relationships is used in developing a function that measures the distance estimated responses from the target values. We choose the optimal condition that minimizes this measure. Also, under the different degree of importance two step procedures are proposed.

• Journal of Integrative Natural Science
• /
• v.13 no.2
• /
• pp.69-75
• /
• 2020

Multiple Input Multiple Output (MIMO) is one of the important wireless communication technologies. This paper proposes MIMO system capacity enhancement by using convolution of periodic circulating vector signals. This signal represents statistical dependencies between transmission signal with discrete noise and receiver signal with the linear shifting of MIMO channel capacity by positive extents. We examine the channel capacity, outage probability and SNR of MIMO receiver by adding log determinant signal with validated in terms of numerical simulation.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.7
• /
• pp.2614-2632
• /
• 2015

A multiple symbol detection (MSD) algorithm is proposed relying on soft information for ultra-wideband systems, where differential space-time block code is employed. The proposed algorithm aims to calculate a posteriori probabilities (APP) of information symbols, where a forward and backward message passing mechanism is implemented based on the BCJR algorithm. Specifically, an MSD metric is analyzed and performed for serving the APP model. Furthermore, an autocorrelation sampling is employed to exploit signals dependencies among different symbols, where the observation window slides one symbol each time. With the aid of the bidirectional message passing mechanism and the proposed sampling approach, the proposed MSD algorithm achieves a better detection performance as compared with the existing MSD. In addition, when the proposed MSD is exploited in conjunction with channel decoding, an iterative soft-input soft-output MSD approach is obtained. Finally, simulations demonstrate that the proposed approaches improve detection performance significantly.

• Journal of Korea Society of Industrial Information Systems
• /
• v.4 no.2
• /
• pp.44-50
• /
• 1999

This paper presents a set of techniques to automatically find rate optimal or near rate optimal implementation of shift-invariant flow graphs on pipelined processor, in which pipeline processor has multiple accumulators and internal datapaths. In such case, the problem to be addressed is the scheduling of multiple instruction streams which control all of the pipeline stages. The goal of an automatic scheduler in this context is to rearrange the order of instructions such that they are executed with minimum iteration period between successive iteration of defining flow graphs. The scheduling algorithm described in this paper also focuses on the problem of removing the hazards due to inter-instruction dependencies.

• Journal of KIISE:Software and Applications
• /
• v.27 no.8
• /
• pp.855-863
• /
• 2000

Although Behavior-Knowledge Space (BKS) method, one of well known decision combination methods, does not need any assumptions in combining the multiple experts, it should theoretically build exponential storage spaces for storing and managing jointly observed K decisions from K experts. That is, combining K experts needs a (K+1)st-order probability distribution. However, it is well known that the distribution becomes unmanageable in storing and estimating, even for a small K. In order to overcome such weakness, it has been studied to decompose a probability distribution into a number of component distributions and to approximate the distribution with a product of the component distributions. One of such previous works is to apply a conditional independence assumption to the distribution. Another work is to approximate the distribution with a product of only first-order tree dependencies or second-order distributions as shown in [1]. In this paper, higher order dependency than the first-order is considered in approximating the distribution and a dependency-based framework is proposed to optimally approximate the (K+1)st-order probability distribution with a product set of dth-order dependencies where ($1{\le}d{\le}K$), and to combine multiple experts based on the product set using the Bayesian formalism. This framework was experimented and evaluated with a standardized CENPARMI data base.