Vol. 50, No. 10,
Oct. 2023
Digital Library
Extracting Instruction Set Architecture Semantics from a Processor Register-transfer Level
http://doi.org/10.5626/JOK.2023.50.10.827
Domain-specific processors have specialized instructions tailored for frequently used operations in a particular domain, which enables them to achieve higher performance. This presents a challenge for program analysis, as the specialized instructions make it difficult to formally describe the instruction semantics. To address this, we present SemTracter, a tool that automatically extracts instruction semantics from a processor implemented in a hardware description language (HDL) at the register-transfer level (RTL). SemTracter obtains the semantics by simulating the processor RTL symbolically and compiling the results into formal instruction semantics using the Sail language. Our evaluation of the SemTracter on a small RISC-V processor RTL showed that it was able to extract the semantics of basic instructions from a 5-stage processor. Most of the RISC-V 32-bit integer base user-level ISA (RV32I) instructions were extracted and the generated semantics matched the manually written version.
A Data Imbalance Minimization Strategy for Scalable Deep Learning Training
Sanha Maeng, Euhyun Moon, Sungyong Park
http://doi.org/10.5626/JOK.2023.50.10.836
As deep neural network training is compute-intensive and takes a very long time, distributed training using clusters with multiple graphics processing units (GPUs) has been widely adopted. The distributed training of deep neural networks is severely slowed due to straggler, i.e., the slowest worker. Hence, previous studies have proposed solutions to the straggler problem. The existing approaches assume that all data samples, such as images, have a constant size, and they do not recognize data imbalance issues, caused by data samples with different sizes, such as videos and audios, while solving the straggler problem. In this paper, we propose a data imbalance minimization (DIM) strategy that considers data imbalance problems to solve the straggler problem caused by imbalanced data samples. Our evaluation on eight NVIDIA Tesla T4 GPUs shows that DIM outperforms the state-of-the-art systems by up to 1.77x speedup with comparable scalability.
Domain Generalized Fashion Object Detection using Style Augmentation and Attention
http://doi.org/10.5626/JOK.2023.50.10.845
With the combination of fashion and computer vision, fashion object detection using deep learning has gained much interest. However, due to the nature of supervision, the performance of the model drops when images with different characteristics are used. We define the dataset with different characteristics and the characteristic of the domain as ‘domain’ and ‘style’, respectively, and propose a new augmentation method that mixes up the existing domain’s style to make a new style. We also use an attention method to extract important features from the images. Using a stylized fashion detection dataset, style deepfashion2, we show that the proposed method enhances performance within all domains.
Improved Open-Domain Conversation Generative Model via Denoising Training of Guide Responses
Bitna Keum, Hongjin Kim, Jinxia Huang, Ohwoog Kwon, Harksoo Kim
http://doi.org/10.5626/JOK.2023.50.10.851
In recent open-domain conversation research, research is actively conducted to combine the strengths of retrieval models and generative models while overcoming their respective weaknesses. However, there is a problem where the generative model either disregards the retrieved response or copies the retrieved response as it is to generate a response. In this paper, we propose a method of mitigating the aforementioned problems. To alleviate the former problem, we filter the retrieved responses and use the gold response together. To address the latter problem, we perform noising on the gold response and the retrieved responses. The generative model enhances the ability to generate responses via denoising training. The effectiveness of our proposed method is verified through human and automatic evaluation.
Effective Robustness Improvement in Medical Image Segmentation : Adversarial Noise Removal by the Input Transform Method
http://doi.org/10.5626/JOK.2023.50.10.859
Adversarial attacks induce the model to make misjudgments by adding fine noise to the deep learning model input data. Deep learning in medical images raises the expectations for computer-assisted diagnosis, but there is a risk of being vulnerable to adversarial attacks. In addition, in the case of the double segmentation model, the defense of adversarial attacks is more difficult, but security studies related to this topic have not received attention. In this study, we perform FGSM attacks ony brain tumor segmentation models and employ input image transformation and gradient regularization as defenses against these attacks. The proposed application of JPEG compression and Gaussian filters effectively removes adversarial noise while maintaining performance in the original images. Moreover, the input image transformation method, when compared to the conventional gradient regularization model for achieving robustness, not only exhibits a higher defense performance but also offers the advantage of being applicable without the need for model retraining. Through this research, we identify vulnerabilities in the security of medical artificial intelligence and propose ensuring robustness that can be applied in the preprocessing stage of the model.
Improvement of Prostate Cancer Aggressiveness Prediction Based on the Deep Learning Model Using Size Normalization and Multiple Loss Functions on Multi-parametric MR Images
Yoon Jo Kim, Julip Jung, Sung Il Hwang, Helen Hong
http://doi.org/10.5626/JOK.2023.50.10.866
Prostate cancer is the second most common cancer in men worldwide, and it is essential to predict the aggressiveness of prostate cancer because the recurrence rate and the effectiveness of treatment vary depending on the aggressiveness. This study enhances the information on small tumors by applying size normalization to predict the aggressiveness of prostate cancer in multi-parametric MR imaging. Additionally, we propose the use of multiple loss functions to distinguish tumors with different aggressiveness while having a similar visual appearance. Experimental results show that the proposed model trained with size-normalized ADC maps achieves an accuracy of 76.28%, sensitivity of 76.81%, specificity of 75.86%, and an AUC of 0.77. Moreover, compared to the tumor-centered ADC maps, size-normalized ADC maps demonstrate improved performance in tumors smaller than 1.5 cm, with an accuracy of 76.47%, sensitivity of 90.91%, and specificity of 69.57%, corresponding to a significant improvement of 17.65%, 27.27%, and 13.05% respectively.
R²FID: Joint Reranker in Fusion-In-Decoder for Open Domain Question Answering over Tables
Sung-Min Lee, Eunhwan Park, Daeryong Seo, Donghyeon Jeon, Inho Kang, Seung-Hoon Na
http://doi.org/10.5626/JOK.2023.50.10.874
Open Domain Question Answering is a challenging problem that aims to generate an answer where reference documents relevant to a question are not provided. Considering that the importance of the QA system in structured data such as tables has recently gradually increased, this paper presents a method for table open domain question answering of Korean, focusing on tabular contents appearing in Wikipedia. In addition, we extensively apply the Joint Reranker based Fusion-In-Decoder to address limitations entailed in table retrieval, Resulting methods based on Joint Reranker led to improvements of an EM of 3.36 and a F1-Score of 3.25 over open domain question answering tasks.
Post-training Methods for Improving Korean Document Summarization Model
So-Eon Kim, Seong-Eun Hong, Gyu-Min Park, Choong Seon Hong, Seong-Bae Park
http://doi.org/10.5626/JOK.2023.50.10.882
The document summarization task generates a short summary based on a long document. Recently, a method using a pre-trained model based on a transformer model showed high performance. However, as it was proved that fine-tuning does not train the model optimally due to the learning gap between pre-training and fine-tuning, post-training, which is additional training between pre-training and fine-tuning, was proposed. This paper proposed two post-training methods for Korean document summarization. One was Korean Spacing, which is for learning Korean structure, and the other was First Sentence Masking, which is for learning about document summarization. Experiments proved that the proposed post-training methods were effective as performance improved when the proposed post-training was used compared to when it was not.
Adaptive Database Intrusion Detection based on Michigan-style Deep Learning Classifier System
http://doi.org/10.5626/JOK.2023.50.10.891
In a role-based access control (RBAC) environment, database intrusion detection can be achieved by designing a role classifier for query transactions and determining it as an intrusion when the predicted role differs from the actually performed role. The current query-role classifier design methods utilize deep learning models, but it was difficult to simultaneously achieve high accuracy and incomplete adaptability for changing patterns. To solve this problem, this study proposes a Michigan-style Deep Learning Classifier System (MDLCS). This method applies a divide-and-conquer strategy that divides the input space into patterns and assigns an optimal classifier, combining the evolutionary computation principle of a Michigan-style learning classifier system with a deep learning classifier to adapt and improve detection performance for real-time changing patterns.The proposed MDLCS method provides strong adaptability and robustness compared to existing intrusion detection methods such as anomaly detection, signature-based detection and behavior-based detection. MDLCS was evaluated in a commercial database following the TPC-E schema and achieved a 26.81%p improved detection performance compared to existing methods under real environmental conditions in which new patterns sequentially emerge.
Reinforcement Learning-Based Trajectory Optimization of Solar Panel-Equipped UAV BS for Energy Efficiency
Dong Uk Kim, Choong Seon Hong, Seong Bae Park, Jong Won Choi
http://doi.org/10.5626/JOK.2023.50.10.899
5G and B5G wireless communication systems use new bands, such as millimeter-wave, to meet user requirements. However, these new bands have limitations such as lower diffraction, lower transmittance, and stronger straightness than traditional frequency bands. To address these limitations, a cellular communication paradigm supported by Unmanned Aerial Vehicle (UAV), makes communication services more flexible than existing ground base stations. However, UAVs have limited battery capacity, which affects the life of telecommunications services. To address this problem, this paper considers UAVs equipped with solar panels. Movement toward energy generation and altitude for user data rate maximization due to solar power of UAVs can consume a lot of energy. Energy generation, data rate maximization, and energy consumption have a trade-off relationship. Therefore, in this study, we proposed a system to locate UAVs that could optimize the above trade-off relationship using agents learned using a reinforcement learning algorithm called "Proximal Policy Optimization (PPO)" and compare the system proposed in this paper.
A Quantitative Comparison of LIME and SHAP using Stamp-Based Distance Method on Image Data
http://doi.org/10.5626/JOK.2023.50.10.906
XAI(eXplainable AI), 인공신경망, MNIST, 도장 기반의 distance method, LIME, SHAP Abstract XAI, or eXplainable AI, is a technique used to explain artificial neural networks in a way that can be understood by humans. However, it is difficult to compare explanations and heat maps produced by XAI algorithms numerically as it is unclear how humans interpret them. This presents a challenge in determining which XAI algorithm is the most effective and accurate in providing explanations. Therefore, we introduced a stamp-based distance method to compare several XAI algorithms and identify the most accurate algorithm. The proposed method involves evaluating the quality of explanations generated by XAI algorithms applied to a deep learning model trained to detect the presence of stamps in the MNIST dataset. This evaluation was performed using statistical techniques to determine the effectiveness of each XAI algorithm. This paper evaluated performances of LIME and SHAP algorithms using the distance method, which compared explanations produced by each algorithm. Result revealed that LIME with the Felzenszwalb method provided more effective explanations than other LIME and SHAP algorithms.
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