Vol. 50, No. 7,
Jul. 2023
Digital Library
Operating System Support-Based Prevention Mechanism for Use-After-Free Attacks on the Glibc Memory Allocator
Chanyoung Park, Jaehyu Lee, Daeyeon Kim, Hyungon Moon
http://doi.org/10.5626/JOK.2023.50.7.541
Use-after-free is a longstanding memory safety problem that causes many security-critical software vulnerabilities. The importance of this problem has motivated the development of numerous mitigation and prevention mechanisms. A class of these mechanisms mimics garbage collectors to prevent use-after-free. The mechanisms delay freeing a heap chunk until they verify the absence of dangling pointers to the chunk. An earlier work, MarkUs, has demonstrated that this delayed-free approach could be implemented with relatively low overhead on many benchmarks. We go further in this direction and present MarKern, a delayed-free mechanism for use-after-free prevention backed by the operating system’s kernel-level support. MarkKern discovers limitations caused by existing Mark-and-Sweep approach implemented only at the user level and solves them through kernel-level supports. Moreover, unlike existing approaches, MarkKern supports the glibc(GNU C Library) Allocator. MarKern addresses these problems with the help of kernel-level support, thereby preventing use-after-free for a program running with the glibc malloc with 18.50% overhead in execution time on average(geometric mean).
Analysis and Mitigation of Training Suspension during Dynamic Scaling of a Distributed Machine Learning Cluster
Younghoon Lim, Junyeol Yu, Euiseong Seo
http://doi.org/10.5626/JOK.2023.50.7.550
Dynamic scaling is important for efficient management of GPU cluster resources. While checkpoint-based suspend-and-resume scaling has been widely used, recent frameworks provide a checkpoint-free approach for transferring trained models from the runtime environment to new GPUs. However, this approach interrupts existing training whenever a new GPU is added and inefficiently synchronizes the training state. To address these issues, we propose an overlapping technique to continue learning during scaling, and a parallel broadcast technique that considers the topology of the GPUs while eliminating redundant model state transfer during synchronization. We tested the proposed approach in Elastic Horovod. Our results show that our approach enabled 72.8% of the suspension time during scaling to be utilized for training, and improved the performance of the model state synchronization by up to 31.7%.
Multi-Document Summarization Use Semantic Similarity and Information Quantity of Sentence
Yeon-Soo Lim, Sunggoo Kwon, Bong-Min Kim, Seong-Bae Park
http://doi.org/10.5626/JOK.2023.50.7.561
Document summarization task has recently emerged as an important task in natural language processing because of the need for delivering concise information. However, it is difficult to obtain a suitable multi-document summarization dataset. In this paper, rather than training with a multi-document summarization dataset, we propose to use a single-document summarization dataset. That is, we propose a multi-document summarization model which generates multiple single-document summaries with a single-document summarization model and then post-processes these summaries. The proposed model consists of three modules: a summary module, a similarity module, and an information module. When multiple documents are entered into the proposed model, the summary module generates summaries of every single document. The similarity module clusters similar summaries by measuring semantic similarity. The information module selects the most informative summary from each similar summary group and collects selected summaries for the final multi-document summary. Experimental results show that the proposed model outperforms the baseline models and it can generate a high-quality multi-document summary. In addition, the performances of each module also show meaningful results.
CLS Token Additional Embedding Method Using GASF and CNN for Transformer based Time Series Data Classification Tasks
Jaejin Seo, Sangwon Lee, Wonik Choi
http://doi.org/10.5626/JOK.2023.50.7.573
Time series data refer to a sequentially determined data set collected for a certain period of time. They are used for prediction, classification, and outlier detection. Although existing artificial intelligence models in the field of time series are mainly based on the Recurrent Neural Network, recent research trends are changing to transformer based models. Although these transformer based models show good performance for time series data prediction problem, they show relatively insufficient performance for classification tasks. In this paper, we propose an embedding method to add special classification tokens generated using Gramian Angular Summation Field and Convolution Neural Network to utilize time series data as input to transformers and found that we could leverage the pre-trained method to improve performance. To show the efficacy of our method, we conducted extensive experiments with 12 different models using the University of California, Riverside dataset. Experimental results show that our proposed model improved the average accuracy of 85 datasets from 1.4% to up to 21.1%.
Zero-Shot Solar Power Efficiency Prediction Method Considering PCC-Based Climate Similarity
Dongjun Kim, Sungwoo Park, Jaeuk Moon, Eenjun Hwang
http://doi.org/10.5626/JOK.2023.50.7.581
Thermal power generation is a power generation method that occupies a large proportion in Korea and abroad due to its low unit price. However, due to its disadvantage of emitting large amounts of harmful substances that can cause health and environmental problems, renewable energy is in the spotlight as an alternative power source. Among various renewable energy generation methods, solar power generation is receiving the most attention because of its advantages such as ease in maintenance. Various solar power generation forecasting studies are being conducted to improve the uncertainty of volatile solar power generation and ensure stability in power supply. However, existing studies have limitations in that they are only applicable when there is a sufficient amount of historical power generation data. Therefore, this paper proposes a solar power generation efficiency prediction method based on zero-shot learning that utilizes historical data of similar regions by concerning weather similarity to solve the cold-start problem, a problem that occurs in prediction when historical data in the target region are lacking. Comparison results revealed that the proposed method had better performance overall in the target area, with a one-hour-based method showing the best prediction performance among other criteria.
Pruning Deep Neural Networks Neurons for Improved Robustness against Adversarial Examples
Gyumin Lim, Gihyuk Ko, Suyoung Lee, Sooel Son
http://doi.org/10.5626/JOK.2023.50.7.588
Deep Neural Networks (DNNs) have a security vulnerability to adversarial examples, which can result in incorrect classification of the DNNs results. In this paper, we assume that the activation patterns of DNNs will differ between normal data and adversarial examples. We propose a revision that prunes neurons that are activated only in the adversarial examples but not in the normal data, by identifying such neurons in the DNNs. We conducted adversarial revision using various adversarial examples generation techniques and used MNIST and CIFAR-10 datasets. The DNNs neurons that were pruned using the MNIST datasets achieved adversarial revision performance that increased up to 100% and 70.20% depending on the pruning method (label-wise and all-label pruning) while maintaining classification accuracy of normal data at above 99%. In contrast, the CIFAR-10 datasets showed a decreased classification accuracy for normal data, but the adversarial revision performance increased up to 99.37% and 47.61% depending on the pruning method. In addition, the efficiency of the proposed pruning-based adversarial revision performance was confirmed through a comparative analysis with adversarial training methods.
Biometrics Performance Improvement of Face Recognition Smart Door Using Binary Classifier
Taeseong Kim, Changsoo Eun, Jongwon Park
http://doi.org/10.5626/JOK.2023.50.7.598
Face recognition based smart door is a biometric system that collects images using a camera and decides whether a visitor is registered by recognizing the face. Recently, with increasing number of single-person households, demand for access convenience has increased. Accordingly, research on smart doors using face recognition method is active. Face recognition based smart doors use deep learning method to recognize visitor"s faces. Difference between the visitor"s face and the registrant"s face is converted into a distance through encoding. If the distance between the two faces is less than the threshold value, the door is opened as it is determined to be the same person. Facial similarity thresholds differ according to region, race, and clothing cultures. Also, biometrics performance varies according to threshold settings. In previous studies, a constant of 0.4 was used as the facial similarity threshold, which was the criterion for determining registration. In this paper, facial similarity thresholds were calculated using five binary classifiers and biometric performance was compared. As a result of the experiment using the LFW dataset, the average EER was improved by 16.59% compared to that when the constant was used.
Multi-task Learning Based Re-ranker for External Knowledge Retrieval in Document-grounded Dialogue Systems
http://doi.org/10.5626/JOK.2023.50.7.606
Document-grounded dialogue systems retrieve external passages related to the dialogue and use them to generate an appropriate response to the user"s utterance. However, the retriever based on the dual-encoder architecture records low performance in finding relevant passages, and the re-ranker to complement the retriever is not sufficiently optimized. In this paper, to solve these problems and perform effective external passage retrieval, we propose a re-ranker based on multi-task learning. The proposed model is a cross-encoder structure that simultaneously learns contrastive learning-based ranking, Masked Language Model (MLM), and Posterior Differential Regularization (PDR) in the fine-tuning stage, enhancing language understanding ability and robustness of the model through auxiliary tasks of MLM and PDR. Evaluation results on the Multidoc2dial dataset show that the proposed model outperforms the baseline model in Recall@1, Recall@5, and Recall@10.
Mini-Batching with Similar-Length Sentences to Quickly Train NMT Models
Daniela N. Rim, Richard Kimera, Heeyoul Choi
http://doi.org/10.5626/JOK.2023.50.7.614
The Transformer model has revolutionized Natural Language Processing tasks such as Neural Machine Translation. Many efforts have been made to study the Transformer architecture to increase its efficiency and accuracy. One potential area for improvement is to address the computation of empty tokens that the Transformer computes only to discard them later, leading to an unnecessary computational burden. To tackle this, we propose an algorithm that sorts translation sentence pairs based on their length before batching and mini-batch with similar-length sentences, which minimizes the waste of computing power. Since the amount of sorting could violate the independent and identically distributed (i.i.d) data assumption, we sort the data partially. In experiments, we apply the proposed method to English-Korean and English-Luganda language pairs for machine translation and show that there are gains in computational time while maintaining the performance. Our method is independent of architectures, so that it can be easily integrated into any training process with flexible data lengths.
A Proposal of Preferred Parent Change Technique in RPL Network Based on Periodic Link Quality Measurement
Hyungtaek Shin, Yubin Ha, Sanghwa Chung
http://doi.org/10.5626/JOK.2023.50.7.621
IEEE 802.15.4e TSCH is a standard improvement of the MAC layer in IEEE 802.15.4 standard for industrial wireless sensor networks. It constitutes a network with high reliability required for industrial environments and Routing Protocol for Low Power and Lossy Networks (RPL). However, the current RPL evaluates the link quality only for neighbors in communication state, makes an exclusion in link quality evaluation for neighbor nodes whose route is not established, consequently, it causes a side effect of biasing the selection of an already established route. Therefore, in this paper, we introduce a technique that enables link quality evaluation for all neighbor nodes with low overhead using Neighbor Link Quality Estimation (NLQE) packets. The proposed method makes it possible to choose an efficient path selection, and simulation results show that the successful delivery rate of packets increased by more than 15% on average and the latency rate decreased by 20% on average compared to the previous one.
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