CSDL Bài trích Báo - Tạp chí
chủ đề: Computer science
1 In-order transition-based parsing for Vietnamese / John Bauer, Hung Bui, Vy Thai, Christopher D. Manning // .- 2023 .- Tập 39 - Số 3 .- P. 207-221 .- 005
In this paper, we implement a general neural constituency parser based on an in-order parser. We apply this parser to the VLSP 2022 Vietnamese treebank, obtaining a test score of .8393 F1, top of the private test leaderboard. Earlier versions of the parser for languages other than Vietnamese have already been included in the publicly released Python package Stanza [35 ]. The next Stanza release will include the Vietnamese model, along with all of the code used in this project.
2 EVJVQA challenge: multilingual visual question answering / Ngan Luu-Thuy Nguyen, Nghia Hieu Nguyen, Duong T.D. Vo, Khanh Quoc Tran, Kiet Van Nguyen // .- 2023 .- Tập 39 - Số 3 .- P. 237-258 .- 005
In this article, we present details of the organization of the challenge, an overview of the methods employed by shared-task participants, and the results. The highest performances are 0.4392 in F1-score and 0.4009 in BLUE on the private test set. The multilingual QA systems proposed by the top 2 teams use ViT for the pre-trained vision model and mT5 for the pre-trained language model, a powerful pre-trained language model based on the transformer architecture. EVJVQA is a challenging dataset that motivates NLP and CV researchers to further explore the multilingual models or systems for visual question answering systems.
3 A new information theory based algorithm for clustering categorical data / Do Si Truong, Lam Thanh Hien, Nguyen Thanh Tung // .- 2023 .- Tập 39 - Số 3 .- P. 259-278 .- 005
In this paper, we review two baseline algorithms for use with categorical data, namely Min-Min Roughness (MMR) and Mean Gain Ratio (MGR), and propose a new algorithm, called Minimum Mean Normalized Variation of Information (MMNVI). MMNVI algorithm uses the Mean Normalized Variation of Information of one attribute concerning another for finding the best clustering attribute, and the entropy of equivalence classes generated by the selected clustering attribute for binary splitting the clustering dataset. Experimental results on real datasets from UCI indicate that the MMNVI algorithm can be used successfully in clustering categorical data. It produces better or equivalent clustering results than the baseline algorithms.
4 Data augmentation analysis of vehicle detection in aerial images / Khang Nguyen // .- 2023 .- Tập 39 - Số 3 .- P. 291-312 .- 005
Drones are increasingly used in various application domains including surveillance, agriculture, delivery, search and rescue missions. Object detection in aerial images (captured by drones) gradually gains more interest in computer vision community. However, research activities are still very few in this area due to numerous challenges such as top-view angle, small-scale object, diverse directions, and data imbalance. In this paper, we investigate different data augmentation techniques. Furthermore, we propose combining data augmentation methods to further enhance the performance of the state-of-the-art object detection methods. Extensive experiments on two datasets, namely, AERIAU, and XDUAV, demonstrate that the combination of random cropped and vertical flipped data boosts the performance of object detectors on aerial images.
5 Parallel fuzzy frequent itemset mining using cellular automata / Trinh T.T. Tran, Thuan T. Nguyen, Giang L. Nguyen, Chau N. Truong // Tin học & Điều khiển học .- 2022 .- V.38-N.4 .- P. 293-310 .- 004
This study presents a reinforced techniquefor mining frequent fuzzy sets based on cellular learning automata (CLA). The results demonstratethat frequent set mining can be accomplished with less running time when the proposed method iscompared to iMFFP and NPSFF methods.
6 Empirical study of feature extraction approaches for image captioning in Vietnamese / Khang Nguyen // Tin học & Điều khiển học .- 2022 .- V.38-N.4 .- P. 327-346 .- 005
This study focus on the image captioning problem in Vietnamese. Indetail, an empirical study of grid-based and region-based feature extraction approaches using currentstate-of-the-art object detection methods is investigated to explore the suitable way to represent theimages in the model space. Each feature type represents images, and the image captioning task istrained using the Transformer-based model. The effectiveness of different feature types is exploredon two Vietnamese datasets: UIT-ViIC and VieCap4H, the two standard benchmark datasets. Theexperimental results show crucial insight into the feature extraction task for image captioning inVietnamese.
7 Evolutionary algorithm for task offloading in vehicular fog computing / Do Bao Son, Vu Tri An, Hiep Khac Vo, Pham Vu Minh, Nguyen Quang Phuc, Nguyen Phi Le, Binh Minh Nguyen, Huynh Thi Thanh Binh // Tin học & Điều khiển học .- 2022 .- V.38-N.4 .- P. 347-364 .- 005
Internet of Things technology was introduced to allow many physical devices to connectover the Internet. The data and tasks generated by these devices put pressure on the traditionalcloud due to high resource and latency demand. Vehicular Fog Computing (VFC) is a concept thatutilizes the computational resources integrated into the vehicles to support the processing of end-user-generated tasks. This research first proposes a bag of tasks offloading framework that allowsvehicles to handle multiple tasks and any given time step. We then implement an evolution-basedalgorithm called Time-Cost-aware Task-Node Mapping (TCaTNM) to optimize completion time andoperating costs simultaneously. The proposed algorithm is evaluated on datasets of different tasksand computing node sizes. The results show that our scheduling algorithm can save more than60%ofmonetary cost than the Particle Swarm Optimization (PSO) algorithm with competitive computationtime. Further evaluations also show that our algorithm has a much faster learning rate and can scaleits performance as the number of tasks and computing nodes increases.
8 Fast computation of direct exponentiation to speed up implementation of dynamic block ciphers / Luong Tran Thi // Tin học & Điều khiển học .- 2022 .- V.38-N.4 .- P. 365-375 .- 005
MDS (maximum distance separable) matrices are ones that come from MDS codes that have been studied for a long time in error correcting code theory and have many applications in block ciphers. To improve the security of block ciphers, dynamic block ciphers can be created. Using MDS matrix transformations is a method used to make block ciphers dynamic. Direct exponentiation is a transformation that can be used to generate dynamic MDS matrices to create a dynamic diffusion layer of the block ciphers. However, for cryptographic algorithms that use an MDS matrix as a component of them, the implementation of matrix multiplication is quite expensive, especially when the matrix has a large size. In this paper, the mathematical basis for quick calculation of direct exponentiation of an MDS matrix will be presented. On that basis, it is to suggest how to apply that fast calculation to dynamic algorithms using the direct exponentiation. This result is very meaningful in software implementation for MDS matrices, especially when implementing dynamic block ciphers to increase execution speed.
9 A data-centric deep learning method for pulmonary nodule detection / Chi Cuong Nguyen, Long Giang Nguyen, Giang Son Tran // Tin học & Điều khiển học .- 2022 .- Vol 38(3) .- P. 229-243 .- 005
In this paper, we follow the direction of data-centric approach for lung nodule detection by proposing a data-centric method to improve detection performance of lung nodules on CT scans. Our method takes into account the dataset-specific features (nodule sizes and aspect ratios) to train detection models as well as add more training data from local Vietnamese hospital. We experiment our method on the three widely used object detection networks (Faster R-CNN, YOLOv3 and RetinaNet). The experimental results show that our proposed method improves detection sensitivity of these object detection models up to 4.24%.
10 A hybrid pso-sa scheme for improving accuracy of fuzzy time series forecasting models / Pham Dinh Phong, Nguyen Duc Du, Pham Hoang Hiep, Tran Xuan Thanh // Tin học & Điều khiển học .- 2022 .- Vol 38 .- P. 257-275 .- 005
Many researches focus on optimizing length of intervals in order to improve forecasting accuracies by utilizing various optimization techniques. In the line of that research trend, in this paper, a hybrid particle swarm optimization combined with simulated annealing (PSO-SA) algorithm is proposed to optimize length of intervals to improve forecasting accuracies. The experimental results in comparison with the existing forecasting models show that the proposed forecasting model is an effective forecasting model.