Recent advancements in computer vision and natural language processing were applied to the Visual Question Answering task. Nonetheless, a significant proportion of models exhibiting high accuracy possess extensive architectural components. This has a significant impact on the process of bringing the technology to practical applications such as assistive devices for the blind and visually impaired, and other related fields. Our research focuses on compressing the Visual Question Answering model on the Vietnamese dataset by utilizing the knowledge distillation method. Furthermore, in order to enhance precision, we have also developed a Mixture of ViVQA Experts system that will adapt to each type of question for improving accuracy while increasing only a few parameters and not wasting time retraining the entire system from scratch. With a total of 204M parameters, this approach has reduced the size by 24.51% compared to the original model while only reducing accuracy by 6.59\% on the overall test set. More specifically, we have made accuracy improvements on each question type: "number" increased by 1.35% and "color" increased by 0.48\% compared to our distillation model. The code and pretrained models are available at: anonymous.

Real-world data is continuously produced andcollected quickly; hence, mining association rules are used tofind relationships between item sets. One of these combinedrule mining techniques is Rare Association Rule Mining(RARM), a method that extracts rare, combined rules with lowsupport but high confidence from the database. Additionally,a large amount of time-sensitive data is created in many fields,and combining valid and outdated data leads to low efficiencyin extracting combined rules. This paper explores a miningmethod that increases rare, combined rules on time-sensitivedata to extract common and rare patterns in the database.The primary approach of the research proposes a new treestructure called ILC-tree, inspired by the LC-tree structurepublished in 2022. Specifically, our approach includes a newtree structure and a fast reconstruction method. The proposedapproach reduces execution time and memory consumptionby approximately 2 times compared to the old tree structure.

The goal of this research is to enhance the capabilityof image retrieval systems to understand images moreeffectively. We present a model designed for searching humanobjects (such as pedestrians or persons) within expansiveimage datasets. Our unique approach involves developing animage retrieval system that incorporates attribute learningand the Human Attribute Ontology (HAO). This researchoffers several key contributions: (1) The development of theHuman Attribute Ontology (HAO) which serves as a repositoryfor storing prior knowledge about images. Thanks toits hierarchical structure, this ontology facilitates the reuse ofprior knowledge, optimizing the subsequent stages of attributelearning and image retrieval; (2) The implementation of aConvolutional Neural Network (CNN) to spearhead attributelearning, leveraging the HAO to enhance accuracy; (3) Thecreation of a Human Image Retrieval system that utilizes bothattribute learning and the HAO. Our system delves deeperby understanding images at the attribute level, highlightingthe advantages of harnessing the ontology to reuse existingknowledge. The efficacy of our methodology is validatedthrough experiments on benchmark datasets like PETA andPa100k achieving state-of-the-art results.

The implementation of Software-Defined Network(SDN) across multiple data centers aims to simplify thecontrol and management of networks. However, the increasingpopularity of SDN has also attracted the attention of attackers.To tackle this problem, it is essential to have an intrusiondetection system (IDS) in place, which plays a crucial role incybersecurity by addressing external threats. The advantageof SDN’s centralized nature is that it facilitates the training ofan IDS based on machine learning. However, there is a scarcityof research specifically focused on intrusion detection in SDN.Existing literature often treats SDN intrusion detection assimilar to traditional computer systems and relies on intrusiondatasets generated for those systems. We explore the issueof intrusion detection in SDN using the most recent publicdataset (InSDN). However, InSDN is an imbalanced data set.In this paper, we have recommended a method to balance thedata as well as a method to find the best features to improvethe quality of IDS using Machine Learning. In addition, wealso suggest a method of classifying SDN network traffic andnormal network traffic. At the same time, we also evaluate theefficiency of the SDN system with the load balancing systemand without the load balancing system.