As software vulnerabilities grow in volume and complexity, researchers proposed various Artificial Intelligence (AI)-based approaches to help under-resourced security analysts find, detect, localize vulnerabilities. However, still have spend a huge amount of effort manually fix or repair such vulnerable functions. Recent work an NMT-based Automated Vulnerability Repair, but it is far from perfect due limitations. In this paper, we propose VulRepair, T5-based automated vulnerability approach...

We propose in this paper a novel approach to tackle the problem of mode collapse encountered generative adversarial network (GAN). Our idea is intuitive but proven be very effective, especially addressing some key limitations GAN. In essence, it combines Kullback-Leibler (KL) and reverse KL divergences into unified objective function, thus exploits complementary statistical properties from these effectively diversify estimated density capturing multi-modes. term our method dual discriminator...

Detecting anomalies in surveillance videos has long been an important but unsolved problem. In particular, many existing solutions are overly sensitive to (often ephemeral) visual artifacts the raw video data, resulting false positives and fragmented detection regions. To overcome such sensitivity capture true with semantic significance, one natural idea is seek validation from abstract representations of videos. This paper introduces a framework robust anomaly using multilevel both...

Identifying vulnerabilities in the source code is essential to protect software systems from cyber security attacks. It, however, also a challenging step that requires specialized expertise and representation. To this end, we aim develop general, practical, programming language-independent model capable of running on various codes libraries without difficulty. Therefore, consider vulnerability detection as an inductive text classification problem propose ReGVD, simple yet effective graph...

Knowledge distillation is an attractive approach for learning compact deep neural networks, which learns a lightweight student model by distilling knowledge from complex teacher model. Attention-based specific form of intermediate feature-based that uses attention mechanisms to encourage the better mimic teacher. However, most previous attention-based approaches perform in spatial domain, primarily affects local regions input image. This may not be sufficient when we need capture broader...

Abstract Many Machine Learning(ML)-based approaches have been proposed to automatically detect, localize, and repair software vulnerabilities. While ML-based methods are more effective than program analysis-based vulnerability analysis tools, few integrated into modern Integrated Development Environments (IDEs), hindering practical adoption. To bridge this critical gap, we propose in article AIBugHunter , a novel Learning-based tool for C/C++ languages that is the Visual Studio Code (VS...

Large language models (LLMs) like ChatGPT (i.e., gpt-3.5-turbo and gpt-4) exhibited remarkable advancement in a range of software engineering tasks associated with source code such as review generation. In this paper, we undertake comprehensive study by instructing for four prevalent vulnerability tasks: function line-level prediction, classification, severity estimation, repair. We compare state-of-the-art designed purposes. Through an empirical assessment employing extensive real-world...

Deep learning-based vulnerability prediction approaches are proposed to help under-resourced security practitioners detect vulnerable functions. However, still do not know what type of vulnerabilities correspond a given (aka CWE-ID). Thus, novel approach explain the for is imperative. In this paper, we propose <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">VulExplainer</i> , an vulnerabilities. We represent as classification task. have...

We propose a new approach to train the Generative Adversarial Nets (GANs) with mixture of generators overcome mode collapsing problem. The main intuition is employ multiple generators, instead using single one as in original GAN. idea simple, yet proven be extremely effective at covering diverse data modes, easily overcoming collapse and delivering state-of-the-art results. A minimax formulation able establish among classifier, discriminator, set similar spirit Generators create samples that...

A typical online kernel learning method faces two fundamental issues: the complexity in dealing with a huge number of observed data points (a.k.a curse kernelization) and difficulty parameters, which often assumed to be fixed. Random Fourier feature is recent effective approach address former by approximating shift-invariant function via Bocher's theorem, allows model maintained directly random space fixed dimension, hence size remains constant w.r.t. size. We further introduce this paper...

Identifying vulnerabilities in the source code is essential to protect software systems from cyber security attacks. It, however, also a challenging step that requires specialized expertise and representation. To this end, we aim develop general, practical, programming language-independent model capable of running on various codes libraries without difficulty. Therefore, consider vulnerability detection as an inductive text classification problem propose ReGVD, simple yet effective graph...

Multi-source Domain Adaptation (MSDA) is more practical but challenging than the conventional unsupervised domain adaptation due to involvement of diverse multiple data sources. Two fundamental challenges MSDA are: (i) how deal with diversity in source domains and (ii) cope shift between target domains. In this paper, address first challenge, we propose a theoretical-guaranteed approach combine experts locally trained on its own achieve combined multi-source teacher that globally predicts...

Due to the ubiquity of computer software, software vulnerability detection (SVD) has become crucial in industry and field security. Two significant issues SVD arise when using machine learning, namely: i) how learn automatic features that can help improve predictive performance ii) overcome scarcity labeled vulnerabilities projects require laborious labeling code by security experts. In this paper, we address these two concerns proposing a novel architecture which leverages deep domain...

Recently, automated vulnerability repair approaches have been widely adopted to combat increasing software security issues. In particular, transformer-based encoder-decoder models achieve competitive results. Whereas vulnerable programs may only consist of a few code areas that need repair, existing AVR lack mechanism guiding their model pay more attention during generation. this article, we propose novel framework inspired by the Vision Transformer based for object detection in computer...

We introduce a new model to deal with imbalanced data sets for novelty detection problems where the normal class of training set can be majority or minority class. The key idea is construct an optimal hypersphere such that inside margin between surface this sphere and outside abnormal are as large possible. Depending on specific real application detection, two margins adjusted achieve best true positive false rates. Experimental results number showed proposed provide better performance...

Recently, Neural Topic Models (NTMs) inspired by variational autoencoders have obtained increasingly research interest due to their promising results on text analysis. However, it is usually hard for existing NTMs achieve good document representation and coherent/diverse topics at the same time. Moreover, they often degrade performance severely short documents. The requirement of reparameterisation could also comprise training quality model flexibility. To address these shortcomings, we...

Using the principle of imitation learning and theory optimal transport we propose in this paper a novel model for unsupervised domain adaptation named Teacher Imitation Domain Adaptation with Optimal Transport (TIDOT). Our includes two cooperative agents: teacher student. The former agent is trained to be an expert on labeled data source domain, whilst latter one aims work unlabeled target domain. More specifically, applied quantify total distance between embedded distributions joint space,...

One of the most current challenging problems in Gaussian process regression (GPR) is to handle large-scale datasets and accommodate an online learning setting where data arrive irregularly on fly. In this paper, we introduce a novel model that could scale with massive datasets. Our approach formulated based alternative representation under geometric optimization views, hence termed geometric-based GP (GoGP). We developed theory guarantee good convergence rate our proposed algorithm always...