• Journal of Broadcast Engineering
• /
• v.27 no.7
• /
• pp.1011-1020
• /
• 2022

Recent high-performance panoptic segmentation models are based on transformer architectures. However, transformer-based panoptic segmentation methods are basically slower than convolution-based methods, since the attention mechanism in the transformer requires quadratic complexity w.r.t. image resolution. Also, sine and cosine computation for positional embedding in the transformer also yields a bottleneck for computation time. To address these problems, we adopt three modules to speed up the inference runtime of the transformer-based panoptic segmentation. First, we perform channel-level reduction using depth-wise separable convolution for inputs of the transformer decoder. Second, we replace sine and cosine-based positional encoding with convolution operations, called conv-embedding. We also apply a separable self-attention to the transformer encoder to lower quadratic complexity to linear one for numbers of image pixels. As result, the proposed model achieves 44% faster frame per second than baseline on ADE20K panoptic validation dataset, when we use all three modules.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.06a
• /
• pp.410-412
• /
• 1998

Subsymbolic Knowledge processing is said to be changed states of networks constructed from small elements. subsymbolic systems also make it possible to use connectionist models for knowledge processing. Connectionist realization such modulus are modulus linked together for solving a given problem. We study using neural networks as distinct actions. The output vectors produced by the neural networks are consider as a new facts. These new facts are then processed to activate another networks or used in the current production rule, The production rule is applying knowledge stored in the knowledge base to make inference. After neural networks knowledge base is constructed and trained. We present a running sample of incorporating neural network knowledge base. We implement using rochester connectionist simulator. We suggest that incorporating neural network knowledge base. Therefore incorporated neural network knowledge base ensures a cleaner solution which results in better perfor s.

• Journal of information and communication convergence engineering
• /
• v.21 no.3
• /
• pp.198-207
• /
• 2023

Responding to changes in artificial intelligence models and the data environment is crucial for increasing data-learning accuracy and inference stability of industrial applications. A learning model that is overfitted to specific training data leads to poor learning performance and a deterioration in flexibility. Therefore, an early stopping technique is used to stop learning at an appropriate time. However, this technique does not consider the homogeneity and independence of the data collected by heterogeneous nodes in a differential network environment, thus resulting in low learning accuracy and degradation of system performance. In this study, the generalization performance of neural networks is maximized, whereas the effect of the homogeneity of datasets is minimized by achieving an accuracy of 99.7%. This corresponds to a decrease in delay time by a factor of 2.33 and improvement in performance by a factor of 2.5 compared with the conventional method.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.264-267
• /
• 2022

Recently, federated learning (FL) has increased prominence as a viable approach for enhancing user privacy and data security by allowing collaborative multi-party model learning without exchanging sensitive data. Despite this, most present FL systems still depend on a centralized aggregator to generate a global model by gathering all submitted models from users, which could expose user privacy and the risk of various threats from malicious users. To solve these issues, we suggested a safe FL framework that employs differential privacy to counter membership inference attacks during the collaborative FL model training process and empowers blockchain to replace the centralized aggregator server.

• Journal of Internet Computing and Services
• /
• v.13 no.2
• /
• pp.87-98
• /
• 2012

Understanding agent's intent is an essential component of the human-computer interaction of ubiquitous computing. Because correct inference of subject's intention in ubiquitous computing system helps particularly to understand situations that involve collaboration among multiple agents or detection of situations that can pose a particular activity. This paper, inspired by people have a mechanism for interpreting one another's actions and for inferring the intentions and goals that underlie action, proposes an approach that allows a computing system to quickly recognize the intent of agents based on experience data acquired through prior capabilities of activities recognition. To proceed intention recognition, proposed method uses formulations of Hidden Markov Models (HMM) to model a system's prior experience and agents' action change, then makes for system infer intents in advance before the agent's actions are finalized while taking the perspective of the agent whose intent should be recognized. Quantitative validation of experimental results, while presenting an accurate rate, an early detection rate and a correct duration rate with detecting the intent of several people performing various activities, shows that proposed research contributes to implement effective intent recognition system.

• The Mathematical Education
• /
• v.37 no.1
• /
• pp.1-13
• /
• 1998

In this paper, we have represented the efficient way how to enumerate the optimal number-right scores to adjust the item difficulty and to improve item discrimination. To estimate the optimal number-right scores in two equivalent math-tests by linear score equating a measurement error model was applied to the true scores observed from a pair of equivalent math-tests assumed to measure same trait. The model specification for true scores which is represented by the bivariate model is a simple regression model to inference the optimal number-right scores and we assume again that the two simple regression lines of raw scores and true scores are independent each other in their error models. We enumerated the difference between mean value of $\chi$＊ and ${\mu}$$\_$$\chi$/ and the difference between the mean value of y＊and a＋b${\mu}$$\_$$\chi$/ by making an inference the estimates from 2 error variable regression model. Furthermore, so as to distinguish from the original score points, the estimated number-right scores y’$\^$*/ as the estimated regression values of true scores with the same coordinate were moved to center points that were composed of such difference values with result of such parallel score moving procedure as above mentioned. We got the asymptotically normal distribution in Figure 5 that was represented as the optimal distribution of the optimal number-right scores so that we could decide the optimal proportion of number-right score in each item. Also by assumption that equivalence of two tests is closely connected to unidimensionality of a student’s ability. we introduce new definition of trait score to evaluate such ability in each item. In this study there are much limitations in getting the real true scores and in analyzing data of the bivariate error model. However, even with these limitations we believe that this study indicates that the estimation of optimal number right scores by using this enumeration procedure could be easily achieved.

• Journal of the Korean Society of Safety
• /
• v.31 no.5
• /
• pp.28-34
• /
• 2016

When a electric power cable is installed in the air for newly or exchanged working, it is necessary at all time to hold a constant tension for an overhead power cable. And also a pendanted power cable is an extreme job to have work in a high sky. For this reason, the objective of this paper developments working robot for preventing disaster that tension of cable installed automatically power cable to hold a constant. So the working robot works at all the time two tasks for mobil and tension it come into a inference between two tasks, control is difficult. Control methode needs to suppress inference of two tasks. In this paper, for installation of overhead power cable, the scheme and control methode of working robot is presented. the robot work at a same time tow tasks that have hold a constant tension of the power cable and move a constant place while unfasten a winding cable at a drum on a chassis. Working robot consist of three parts with mobile system, tension system and control part. As it is applied the feedback/feedforward control, methode of hybrid control is established to suppress that interference come into between two tasks. The simulation programs is made out using models of mobil and tension system, and a proposed controllers. In accordance with simulation, the model of each systems is discussed to make out proper. And also parameters of controllers is selected a suitable value and the driving performance of robot is evaluated.

• Journal of Intelligence and Information Systems
• /
• v.15 no.4
• /
• pp.123-140
• /
• 2009

Fuzzy neural network is an integrated model of artificial neural network and fuzzy system and it has been successfully applied in control and forecasting area. Recently ANFIS(Adaptive Network-based Fuzzy Inference System) has been noticed widely among various fuzzy neural network models because of outstanding performance of control and forecasting accuracy. ANFIS has capability to refine its fuzzy rules interactively with human expert. In particular, when we use initial rule structure for machine learning which is generated from human expert, it is highly probable to reach global optimum solution as well as shorten time to convergence. We propose metrics to evaluate interpretability of generated rules as a means of acquiring domain knowledge and compare level of interpretability of ANFIS fuzzy rules to those of C5.0 classification rules. The proposed metrics also can be used to evaluate capability of rule generation for the various machine learning methods.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.1
• /
• pp.81-86
• /
• 2005

In this study, we introduce a new category of fuzzy inference systems based on information granulation to carry out the model identification of complex and nonlinear systems. Informally speaking, information granules are viewed as linked collections of objects (data, in particular) drawn together by the criteria of proximity, similarity, or functionality Granulation of information with the aid of Hard C-Means (HCM) clustering help determine the initial parameters of fuzzy model such as the initial apexes of the membership functions and the initial values of polynomial functions being used in the premise and consequence part of the fuzzy rules. And the initial parameters are tuned effectively with the aid of the genetic algorithms(GAs) and the least square method (LSM). An aggregate objective function with a weighting factor is also used in order to achieve a balance between performance of the fuzzy model. The proposed model is evaluated with using a numerical example and is contrasted with the performance of conventional fuzzy models in the literature.

• The Korean Journal of Applied Statistics
• /
• v.29 no.6
• /
• pp.1107-1116
• /
• 2016

Recent astronomical survey observations have produced substantial amounts of data as well as completely changed conventional methods of analyzing astronomical data. Both classical statistical inference and modern machine learning methods have been used in every step of data analysis that range from data calibration to inferences of physical models. We are seeing the growing popularity of using machine learning methods in classical problems of astronomical data analysis due to low-cost data acquisition using cheap large-scale detectors and fast computer networks that enable us to share large volumes of data. It is common to consider the effects of inhomogeneous spatial and temporal coverage in the analysis of big astronomical data. The growing size of the data requires us to use parallel distributed computing environments as well as machine learning algorithms. Distributed data analysis systems have not been adopted widely for the general analysis of massive astronomical data. Gathering adequate training data is expensive in observation and learning data are generally collected from multiple data sources in astronomy; therefore, semi-supervised and ensemble machine learning methods will become important for the analysis of big astronomical data.