Vol. 49, No. 12,
Dec. 2022
Digital Library
Design and Implemention of Time-Triggered Architecture for Multicore Automotive Systems
http://doi.org/10.5626/JOK.2022.49.12.1043
Recently, automotive electrical/electronic (E/E) architectures have considered the Multicore AUTOSAR platform for guaranteeing the safety and performance of automotive systems. However, inter-core communication response time delays due to spinning caused by spinlock deteriorate Multicore performance. This paper presents the design of a Time-Triggered Architecture (TTA) to optimize the Multicore system. In our approach, we present the TTA design methodology, including the task allocation algorithm using DQN reinforcement for inter-core load balancing, the Harmonic-Period setting algorithm, and the task Offset, Deadline setting algorithm. Then, we proposed a Timing Violation detection method using Data Version to apply it to the AUTOSAR platform. For verification, we applied the TTA algorithm to the Fuel Cell Controller (FCU) task model. Our simulations showed that the load balancing rate was improved by 94% compared to the existing controller, and its scalability covered at least 78% of the optimal value. It also showed that mutual exclusion was enforced and confirmed that each algorithm was well applied.
Copy-Paste Based Image Data Augmentation Method Using
http://doi.org/10.5626/JOK.2022.49.12.1056
In the field of computer vision, massive well-annotated image data are essential to achieve good performance of a convolutional neural network (CNN) model. However, in real world applications, gathering massive well-annotated data is a difficult and time-consuming job. Thus, image data augmentation has been continually studied. In this paper, we proposed an image data augmentation method that could generate more diverse image data by combining generative adversarial network (GAN) and copy-paste based augmentation. The proposed method generated not pixel-level or image-level augmentation, but object-level augmentation by cutting off segmentation boundaries(mask) instead of bounding boxes. It then applyied GAN to transform objects.
An Energy-Efficient HVAC Control Scheme Based on Deep Reinforcement Learning Using Liquefied Natural Gas Carrier Environment Prediction Model
Youngeun Chae, Jaeseong Kim, Jin-Sung Ok, Young-Kyoon Suh
http://doi.org/10.5626/JOK.2022.49.12.1062
This paper proposes a heating, ventilation, and air conditioning (HVAC) control scheme based on deep reinforcement learning to stably maintain the internal environment of an LNG cargo hold under construction and minimize energy consumption. Since a particular environment such as inside of a cargo ship under construction is readily influenced by several factors, it is difficult to accurately forecast indoor temperature and humidity information and automatically control HVAC system. To alleviate this problem, we propose a novel scheme to steadily controlling an indoor environment via an HVAC control agent trained through a deep reinforcement learning model. In this scheme, we construct an indoor-environment state prediction model based on correlational analyses of collected data without expertise concerning the operating circumstance, define the state and action based on the model, and then build the agent trained with a policy through a reward function. To assess the validity of the proposed scheme, we conduct HVAC control performance evaluation in a simulated environment built using the data collected from an actual LNGC HVAC system. Our simulation results show that the Double Deep Q-Network (DQN) model was the most effective for HVAC control among three types of reinforcement learning models that we considered in this study. Also, the results reveal that the trained agent could reduce average daily power consumption by 28.2% while stabilizing indoor environment of the cargo hold within user-specified temperature range.
BERT-based Two-Stage Classification Models and Co-Attention Mechanism for Diagnosing Dementia and Schizophrenia-related Disease
Min-Kyo Jung, Seung-Hoon Na, Ko Woon Kim, Byoung-Soo Shin, Young-Chul Chung
http://doi.org/10.5626/JOK.2022.49.12.1071
Noting the recently increasing number of patients, we present deep learning methods for automatically diagnosing dementia and schizophrenia by exploring the use of the novel two-stage classification and the co-attention mechanism. First, the two-stage classification consists of two steps-the perplexity-based classification and the standard BERT-based classification. 1) the perplexity-based classification first prepares two types of BERTs, i.e., control-specific and patients-specific BERTs, pretrained from transcripts for controls and patients as the additional pretraining datasets, respectively, and then performs a simple threshold-based classification based on the difference between perplexity values of two BERTs for an input test transcript; then, for ambiguous cases where the perplexity difference only does not provide sufficient evidence for the classification, the standard BERT-based classification is performed based on a fine-tuned BERT. Second, the co-attention mechanism enriches the BERT-based representations from a doctor’s transcript and a client’s one by applying the cross-attention over them using the shared affinity matrix, and performs the classification based on the enriched co-attentive representations. Experiment results on a large-scale dataset of Korean transcripts show that the proposed two-stage classification outperforms the baseline BERT model on 4 out of 7 subtasks and the use of the co-attention mechanism achieves the best F1 score for 4 out of 8 subtasks.
Incremental and Automated Mock Generation using Execution Logs for the Abstraction of Embedded Software Modules
http://doi.org/10.5626/JOK.2022.49.12.1088
One of the barriers to applying verification methods, such as dynamic tests, static analysis, and model checking, to unit modules is constructing the verification environment, including external components developed by third parties. Current practices replace such external components with manual stubs. This requires time and effort and suffers from inaccuracy of the stubs, resulting in low verification efficiency. This work proposes a log-based automated mock generation approach using the Program-By-Example (PBE) synthesis method and performed experimental analysis of its efficiency and accuracy. The suggested method is an incremental mock synthesis method under the condition that the source code or the specification of the target component does not exist. Through a set of experiments, this study demonstrated that log-based incremental mock generation could achieve accuracy and efficiency comparable to the ideal case where oracles exist.
Utilizing External Knowledge in Natural Language Video Localization
Daneul Kim, Daechul Ahn, Jonghyun Choi
http://doi.org/10.5626/JOK.2022.49.12.1097
State-of-the-art Natural Language Video Localization (NLVL) models mostly use existing labels to train. The use of either full-supervision or weak-supervision needs costly annotations, which are not applicable to the real-world NLVL problems. Thus, in this study, we propose the framework of External Knowledge-based Natural Language Video Localization (EK-NLVL), which leverages the idea of generating the pseudo-supervision based on a captioning model that generates sentences from the given frames and summarizes them to ground the video event. Moreover, we propose data augmentation using the pre-trained multi-modal representation learning model CLIP for visual-aligned sentence filtering to generate pseudo-sentences that could effectively provide better quality augmentation. We also propose a new model, Query-Attentive on Segmentations Network (QAS) which effectively uses external knowledge for the NLVL task. Experiments using the Charades-STA dataset demonstrated the efficacy of our method compared to the existing models.
FedGC: Global Consistency Regularization for Federated Semi-supervised Learning
http://doi.org/10.5626/JOK.2022.49.12.1108
Recently, in the field of artificial intelligence, methods of learning neural network models in distributed environments that use sufficient data and hardware have been actively studied. Among them, federated learning, which guarantees privacy preservation without sharing data, has been a dominant scheme. However, existing federated learning methods assume supervised learning using only labeled data. Since labeling costs are incurred for supervised learning, the assumption that only label data exists in the clients is unrealistic. Therefore, this study proposes a federated semi-supervised learning method using both labeled data and unlabeled data, considering a more realistic situation where only labeled data exists on the server and unlabeled data on the client. We designed a loss function considering consistency regularization between the output distributions of the server and client models and analyzed how to adjust the influence of consistency regularization. The proposed method improved the performance of existing semi-supervised learning methods in federated learning settings, and through additional experiments, we analyzed the influence of the loss term and verified the validity of the proposed method.
Re-Generation of Models via Generative Adversarial Networks and Bayesian Neural Networks for Task-Incremental Learning
http://doi.org/10.5626/JOK.2022.49.12.1115
In contrast to the human ability of continual learning, deep learning models have considerable difficulty maintaining their original performance when the model learns a series of incrementally arriving tasks. In this paper, we propose ParameterGAN, a novel task-incremental learning approach based on model synthesis. The proposed method leverages adversarial generative learning to regenerate neural networks themselves which have a parameter distribution similar to that of a pre-trained Bayesian network. Also, using pseudo-rehearsal methods, ParameterGAN enables continual learning by regenerating the networks of all previous tasks without catastrophic forgetting. Our experiment showed that the accuracy of the synthetic model composed of regenerated parameters was comparable to that of the pre-trained model, and the proposed method outperformed the other SOTA methods in the comparative experiments using the popular task-incremental learning benchmarks Split-MNIST and Permuted-MNIST.
A Neuro Symbolic Ensemble Language Representation Using Syntactic and Semantic Information
http://doi.org/10.5626/JOK.2022.49.12.1124
With the advent of the giant language model, natural language processing has presented an opportunity to break through the wall. However, since super-large language models only use information extracted from the context, they chose to simply increase the size of the model or the amount of data to improve performance. This approach increases the resources consumed by the language model. In this paper, we propose a Neuro Symbolic Ensemble Language Representation (NSELR) that learns the semantic information of vocabulary together with a language model that uses only contextual information. This model uses the semantic constraint information of hypernym and verb-noun relation in the Korean WordNet (UWordNet) and additionally uses the semantic vectors of words. The NSELR was tested in four domains, and it showed better performance than the existing model in the machine reading comprehension. In addition, the speed of learning convergence was faster than that of the existing model, and when there was insufficient data in the application area, it showed better performance than the existing model.
Generating Counterfactual Examples through Generating Adversarial Examples
http://doi.org/10.5626/JOK.2022.49.12.1132
The advance of artificial intelligence (AI) has brought numerous conveniences. However, the complex structure of AI models makes it challenging to understand the inner working of AI. Counterfactual explanation is a method using counterfactual examples, in which minimum perceptible perturbations are applied to change classification results, to explain AI. Adversarial examples are data modified for causing AI models to misclassify the data. Unlike counterfactual examples, perturbations applied to adversarial examples are difficult for humans to perceive. In a simple model, generating adversarial examples is similar to generating counterfactual examples. In contrast, it is different in deep learning because the cognitive difference between humans and deep learning models is often huge. Nevertheless, we confirmed that adversarial examples generated by certain deep learning models were similar to counterfactual examples. In this paper, we analyzed the structure and conditions of deep learning models in which adversarial examples were similar to counterfactual examples. We also proposed a new metric, partial concentrated change (PCC), and compared adversarial examples generated from different models using existing metrics and the proposed PCC.
Approximating the Accuracy of Classification Models Using Self-differential Testing
Jubin Lee, Taeho Kim, Yu-Seung Ma
http://doi.org/10.5626/JOK.2022.49.12.1143
Differential testing is a traditional software testing technique that detects errors by observing whether similar applications generate different outputs for the same input. Differential testing is also used in artificial intelligence systems. Existing research involves the cost of finding a high-quality reference neural network with the same function as the target neural network but different architectures. We propose a self-differential testing technique that evaluates a classification model by making a reference model using a target neural network without the need to find the neural network of another architecture when differential testing. Experiments confirmed that self-differential testing produced similar effects at a lower cost than the existing research that requires other reference models. In addition, we propose an accuracy approximation method for classification models using self-differential analysis, which is an application of self-differential testing. The approximate accuracy through self-differential testing was confirmed to show a small difference of 0.0002 to 0.09 from the actual accuracy in experiments using similar datasets of MNIST and CIFAR10.
An Efficient Continuous Subgraph Matching Technique for Graph Stream Processing in a Memory-constrained Environment
Somin Lee, Sanghyeuk Kim, Hyeonbyeong Lee, Dojin Choi, Jongtae Lim, Kyoungsoo Bok, Jaesoo Yoo
http://doi.org/10.5626/JOK.2022.49.12.1154
Recently, with the proliferation of social network services, the size of graph data has been becoming increasingly vast and graph data are changed in real-time. Therefore, it is necessary to perform continuous query processing on real-time graph streams. Moreover, it is difficult keep the entire large graph data in the main memory since its size is constrained in real-world application environments. Consequently, continuous subgraph matching techniques are required by considering memory-constrained environments. In this paper, we propose a continuous subgraph matching technique for graph streams in a memory-constrained environment. The proposed technique consists of modules such as index manager, query processor, and cache manager for efficient continuous subgraph matching. We conduct performance evaluations to demonstrate the superiority of the proposed technique.
A Method to Calculate Color Temperature of Natural Light Using a Representative Trend Line
SeungTaek Oh, YunJi Kim, JaeHyun Lim
http://doi.org/10.5626/JOK.2022.49.12.1166
The color temperature cycle of natural light helps in promoting good health, such as maintaining the human circadian rhythm. However, although the current lighting technology provides the color temperature during a specific period or the color temperature calculated from characteristics of natural light, a continuous color temperature cycle appearing on the ground cannot be realized. Therefore, this paper proposes a calculation method of natural light color temperature cycles based on the representative trend line to realize the cyclic properties of the color temperature of natural light. First, the color temperature cycle for the sunny day based on each solar term was extracted through the natural light characteristic DB collected through actual measurement. A representative trend line was created through analysis, and then a mean deviation between the representative trend line and color temperature cycle by each solar term was obtained. After calculating the color temperature cycle from the representative trend line from sunrise to sunset for each solar term, a standard of color temperatures of natural light for the solar term/days through shift calculation based on the mean deviation of the color temperature was calculated. The proposed method proved that an accurate color temperature could be calculated within the mean absolute error of 39.8K.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr