Deep Domain Adaptation for Vulnerable Code Function Identification

Nguyen, Van Du; Le, Trung; Le, Tue; Nguyen, Khanh; deVel, Olivier; Montague, Paul; Qu, Lizhen; Phung, Dinh

doi:10.1109/ijcnn.2019.8851923

Cited by 23 publications

(29 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, most of aforementioned works mainly focus on transfer learning in the computer vision domain. The work of [16] is the first work which applies deep domain adaptation to SVD with promising predictive performance on real-world source code projects. The underlying idea is to employ the GAN to close the gap between the source and target domains in the joint space and enforce the clustering assumption [2] to utilize the information carried in the unlabeled target samples in a semi-supervised context.…”

Section: Introductionmentioning

confidence: 99%

“…We name our mechanism when applied to SVD as Dual Generator-Discriminator Deep Code Domain Adaptation Network (Dual-GD-DDAN). We empirically demonstrate that our Dual-GD-DDAN can overcome the missing mode and boundary distortion problems which is likely to happen as in Deep Code Domain Adaptation (DDAN) [16] in which the GAN was solely applied to close the gap between the source and target domains in the joint space (see the discussion in Sects. 2.3 and 3.3, and the visualization in Fig.…”

Section: Introductionmentioning

confidence: 99%

“…3). In addition, we incorporate the relevant approaches -minimizing the conditional entropy and manifold regularization with spectral graph -proposed in [16] to enforce the clustering assumption [2] and arrive at a new model named Dual Generator-Discriminator Semi-supervised Deep Code Domain Adaptation Network (Dual-GD-SDDAN). We further demonstrate that our Dual-GD-SDDAN can overcome the mode collapsing problem better than SCDAN in [16], hence obtaining better predictive performance.…”

Section: Introductionmentioning

confidence: 99%

“…We conducted experiments using the data sets collected by [13], that consist of five real-world software projects: FFmpeg, LibTIFF, LibPNG, VLC and Pidgin to compare our proposed Dual-GD-DDAN and Dual-GD-SDDAN with the baselines. The baselines consider to include VULD (i.e., the model proposed in [12] without domain adaptation), MMD, DIRT-T, DDAN and SCDAN as mentioned [16] and D2GAN [15] (a variant of the GAN using dual-discriminator to reduce the mode collapse for which we apply this mechanism in the joint feature space). Our experimental results show that our proposed methods are able to overcome the negative impact of the missing mode and boundary distortion problems inherent in deep domain adaptation approaches when solely using the GAN principle as in DDAN and SCDAN [16].…”

Section: Introductionmentioning

confidence: 99%

“…The baselines consider to include VULD (i.e., the model proposed in [12] without domain adaptation), MMD, DIRT-T, DDAN and SCDAN as mentioned [16] and D2GAN [15] (a variant of the GAN using dual-discriminator to reduce the mode collapse for which we apply this mechanism in the joint feature space). Our experimental results show that our proposed methods are able to overcome the negative impact of the missing mode and boundary distortion problems inherent in deep domain adaptation approaches when solely using the GAN principle as in DDAN and SCDAN [16]. In addition, our method outperforms the rival baselines in terms of predictive performance by a wide margin.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Dual-Component Deep Domain Adaptation: A New Approach for Cross Project Software Vulnerability Detection

Nguyen

Vel

et al. 2020

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Owing to the ubiquity of computer software, software vulnerability detection (SVD) has become an important problem in the software industry and computer security. One of the most crucial issues in SVD is coping with the scarcity of labeled vulnerabilities in projects that require the laborious manual labeling of code by software security experts. One possible solution is to employ deep domain adaptation (DA) which has recently witnessed enormous success in transferring learning from structural labeled to unlabeled data sources. Generative adversarial network (GAN) is a technique that attempts to bridge the gap between source and target data in the joint space and emerges as a building block to develop deep DA approaches with state-of-the-art performance. However, deep DA approaches using the GAN principle to close the gap are subject to the mode collapsing problem that negatively impacts the predictive performance. Our aim in this paper is to propose Dual Generator-Discriminator Deep Code Domain Adaptation Network (Dual-GD-DDAN) for tackling the problem of transfer learning from labeled to unlabeled software projects in SVD to resolve the mode collapsing problem faced in previous approaches. The experimental results on real-world software projects show that our method outperforms stateof-the-art baselines by a wide margin.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Dual-Component Deep Domain Adaptation: A New Approach for Cross Project Software Vulnerability Detection

Nguyen

Vel

et al. 2020

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

Code Action Network for Binary Function Scope Identification

Nguyen

Le³

et al. 2020

Advances in Knowledge Discovery and Data Mining

Self Cite

View full text Add to dashboard Cite

Function identification is a preliminary step in binary analysis for many applications from malware detection, common vulnerability detection and binary instrumentation to name a few. In this paper, we propose the Code Action Network (CAN) whose key idea is to encode the task of function scope identification to a sequence of three action states NI (i.e., next inclusion), NE (i.e., next exclusion), and FE (i.e., function end) to efficiently and effectively tackle function scope identification, the hardest and most crucial task in function identification. A bidirectional Recurrent Neural Network is trained to match binary programs with their sequence of action states. To work out function scopes in a binary, this binary is first fed to a trained CAN to output its sequence of action states which can be further decoded to know the function scopes in the binary. We undertake extensive experiments to compare our proposed method with other state-of-the-art baselines. Experimental results demonstrate that our proposed method outperforms the state-of-the-art baselines in terms of predictive performance on real-world datasets which include binaries from well-known libraries.

show abstract