Tọa đàm mang đến cơ hội cho các nhà khoa học, các chuyên gia, sinh viên, nghiên cứu sinh trao đổi với các đồng nghiệp trong nước, quốc tế và tìm khả năng phát triển hợp tác về An toàn thông tin.
|Thứ Ba, ngày 30/10/2018, 08h00-11h00||
Deep Learning For Natural Language Processing And Beyond
|Hội trường K12 – 1916-S1|
|Thứ Sáu, ngày 02/11/2018, 14h00-17h00||
||Hội trường K12 – 1916-S1|
|Chủ Nhật, ngày 04/11/2018, 08h00-17h00||Tọa đàm Khoa học công nghệ Việt Nam – Nhật Bản và Định hướng nghề nghiệp trong lĩnh vực An toàn thông tin||Hội trường S5 - HVKTQS|
TỌA ĐÀM KHOA HỌC CÔNG NGHỆ VIỆT NAM – NHẬT BẢN VÀ HƯỚNG NGHIỆP 2018 CHỦ ĐỀ “SINH VIÊN ATTT BẢN LĨNH, SÁNG TẠO, ĐI ĐẦU TRONG CÁCH MẠNG CÔNG NGHIỆP 4.0”|
Hà nội, ngày 4 tháng 11 năm 2018
|8h00 - 8h30||Đón tiếp đại biểu|
Học viện KTQS
|8h30 - 8h40||Tuyên bố lý do, giới thiệu đại biểu khách mời|
Thông qua nội dung Chương trình
Học viện KTQS
|8h40 - 8h50||Phát biểu khai mạc và chào mừng|
Head of LQDTU Welcome
|PGS. TS Lê Kỳ Nam|
Phó giám đốc Học viện KTQS
|8h50 - 9h15||Sinh viên ATTT – Sẵn sàng thể hiện bản lĩnh trong cuộc Cách mạng công nghiệp 4.0||Ông Nguyễn Huy Dũng|
Phó Cục trưởng Cục An toàn thông tin, Bộ Thông tin và Truyền thông
|9h15 - 9h40||Môi trường ảo đào tạo An ninh mạng|
CyTrONE: Cybersecurity Training Framework
|GS Lim Yuto|
Viện JAIST - Nhật Bản
|9h40 - 10h05||Môi trường thử sức chuyên nghiệp cho sinh viên ATTT||Ông Khổng Huy Hùng|
Giám đốc Công ty cổ phần Công nghệ An ninh không gian mạng Việt Nam - VNCS
|10h05 - 10h30||Cấu trúc cây Byzantine Fault Torelant Merkle và tạo chữ ký số: Ứng dụng trong đăng nhập và giao dịch tài chính an toàn.|
Byzantine Fault Tolerant Merkle Tree Construction and Digital Signature Creation: Their applications to Secure Logging and Financial Transaction Services
|GS Ono Satoshi|
Đại học Kogakuin – Tokyo
|10h30 - 10h55||Sinh viên chung tay xây dựng sản phẩm ATTT thương hiệu Việt||Phó Chủ tịch phụ trách An ninh mạng|
Tập đoàn BKav
|10h55 - 11h20||Ứng dụng kỹ tuật xử lý ngôn ngữ tự nhiên trong phân tích mã độc.|
Formal semantics database for binaries: An approach of systematic generation of dynamic symbolic execution tools on various binaries.
|GS Ogawa |
Viện JAIST - Nhật Bản
|11h20 - 11h30||Trao quà, học bổng||Cục ATTT/ Bộ TTTT|
|11h30||Bế mạc Hội nghị||Khoa CNTT/|
Học viện KTQS
|THE FIRST VIETNAM – JAPAN WORKSHOP ON INFORMATION SECURITY AND INNOVATION EXPO|
Ha noi, 4th November 2018
|Conference room - S4 building|
Chair: Prof. Bui Thu Lam and Prof Tran Nguyen Ngoc
|13h30 - 13h35||Workshop Opening||Prof Bui Thu Lam|
|13h35 - 14h05||Reliable Agent Communication in Logic||Prof Tojo|
JAIST - Japan
|14h05 - 14h35||Deep-learning methods for Intrusion Detection Systems||Prof Tran Nguyen Ngoc|
Dr Nguyen Viet Hung
FIT - LQDTU
|14h35 - 15h05|
Dependable Smart Homes using Cyber-Physical Systems Approach
|Prof Lim Yuto|
JAIST - Japan
|15h05 - 15h35||An improved QR decomposition for color image watermarking in information security||Dr Ta Minh Thanh, Phuong Thi Nha|
FIT - LQDTU
|15h35 - 15h55||Tea/ Coffee Break|
|15h55 - 16h25||Latent representation models for network anomaly detection||Dr Cao Van Loi|
FIT - LQDTU
|16h25- 16h55||Natural language processing for legal engineering and its application||Prof Le Minh Nguyen |
JAIST - Japan
|16h55 - 17h25||Convolutional Neural Networks over Control Flow Graphs for Software Defect Prediction||Dr Anh Việt Phan, |
Prof Minh Le Nguyen,
Prof Lam Thu Bui
FIT - LQDTU
|17h25 - 17h55||Generative Adversarial Networks for Generating Synthesized Attacks in Intrusion Detection System||Dr Nguyen Quang Uy, Vu Thi Ly |
FIT - LQDTU
This talk presents new social infrastructures for secure logging and financial transaction services, which have high performance, reliable and Byzantine Fault Tolerant properties. It also clarifies underlying technologies enabling these goals. This talk comprises two parts: The first part deals with technology elements essential for above mentioned infrastructures, namely Merkle tree construction and digital signature creation, both having Byzantine Fault Tolerance.The second part exemplifies their applications to secure logging and financial transaction service infrastructures.
With LORIA (Nancy) and CEA (Paris), we are starting the project of the formal semantics database for binaries. There are several tools for formal methods on binaries, especially on x86/32. When we develop such tools, extracting formal semantics is a crucial step, which has two problems: (1) the number of instructions is quite large. For instance, the number of x86 instructions is said to be >1000. (2) Human makes errors. Our plan is to overcome such problems, we will approach with the systematic extraction of formal semantics from natural language specification of a various assembly. From Autumn 2013, we have been working on the development of BE-PUM (bepum.jaist.ac.jp) for x86/32, and after 3 years effort, it covers about 200 x86 instructions (among >1000). With such experiences, we further extend the coverage by NLP techniques on natural language specifications collected from the Internet. Our experiences show that about 60% of collected specifications can be automatically handled by simple NLP techniques. We will also mention the recent trial on ARM Cortex-M. Adding to such systematic semntics extraction, we will compose the formal semantics database for binaries by collecting the interpretations given by different cites. We assume that they are in SMT format, then we can automatically check the inconsistency among them by using SMT solvers and make our understanding of binaries more concrete. The start will be x86/32, which is hoped to be extended to other platforms, e.g., ARM and MIPS..
We introduce the dynamic symbolic execution for binaries. Malware often applies obfuscation techniques, which are main reasons of the failure of virus detection of anti-virus software (which is mostly based on machine learning techniques on binaries) and the failure of commercial disassemblers like IDApro and CAPSTONE. We see how the dynamic symbolic execution overcomes such obfuscation techniques. The talk will start with the basis of symbolic execution and how it is adapted to binaries. As an example, we see an application on the packer identification. We mainly focus on x86/32 and further discuss on the extensions on various binaries, e.g., ARM, MIPS, based on a systematic generation of dynamic symbolic execution tools from natural language specification of an assembly language. From Autumn 2013, we have been working on the development of BE-PUM (bepum.jaist.ac.jp) for x86/32, and after 3 years effort, it covers about 200 x86 instructions (among >1000) and 400 Windows APIs (among >4000). With such experiences, we further extend the coverage by NLP techniques on natural language specifications collected from the Internet. Our experiences show that about 60% of collected specifications can be automatically handled by simple NLP techniques..
The Tutorial begins with the basic of feed-forward neural network and relevant fundamental knowledge for deep learning. We then introduce more specialized neural network models, including Convolutional Neural Network, Recurrent Neural Network, and attention-based models. In the second part, we will present how these models and techniques can be applied to some interesting problems of natural language processing including sentiment classification, textual entailment recognition, natural language generation, and question answering. The last part of the tutorial will show how we can adapt deep learning and natural language processing techniques for program analysis..
Our society is regulated by a lot of laws which are related mutually. When a society is viewed as a system, laws can be viewed as the specifications for the society. In the upcoming e-Society, laws have more important roles for achieving a trustworthy society and we expect a methodology which treats a system-oriented aspect of laws. Legal Engineering is the field that studies the methodology and applies information science, software engineering and artificial intelligence to laws for supporting legislation and to implement laws using computers. As laws are written in natural language, natural language processing is essential for Legal Engineering. In this talk, we present our works on natural language processing for Legal Engineering. We also highlight our current deep learning-based techniques for analyzing legal documents and our system participating on the Fifth Competition on Legal Information Extraction/Entailment (COLIEE-2018)..
The dependability is an essential measure of system ability to provide services such as availability, reliability, safety, security, and maintainability. Today, this concept has been widely and usually used in design and analysis of information systems. Due to the rapid increase of home devices and appliances in the smart homes, its importance in control and management has increased drastically in order to guarantee the dependability of smart homes using Cyber-Physical Systems approach. In this talk, the current trend and future promising smart homes will be discussed. The Cyber-Physical Systems is highly expected to offer a dependable and unified network operation and management to facilitate efficient implementation and reduce processing load in the smart home domain.
Existing defects in software components is unavoidable and leads to not only a waste of time and money but also many serious consequences. To build predictive models, traditional methods apply machine learning on handcrafted features or tree representations of programs. However, the performance is not high since the existing features and tree structures often fail to capture the semantics of programs. To explore deeply programs' semantics, this research proposes to leverage precise graphs representing program execution flows, and deep neural networks for automatically learning defect features. Firstly, control flow graphs are constructed from the assembly instructions obtained by compiling source code; we thereafter apply multi-view multi-layer directed graph-based convolutional neural networks (DGCNNs) to learn features. The experiments on four real-world datasets show that our method significantly outperforms the baselines including several other deep learning approaches.
One of the urgent issues in cyber security is the early detection of attacks against information systems. To deal with varied and constantly changing attacks, the current trend is to utilise artificial intelligence methods, especially machine learning, to detect attacks based on anomalies. Machine learning can provide IDS methods to detect current, new and subtle attacks without extensive human-based training or intervention. Deep-learning is a class of machine learning techniques that has many layers of information processing stages in hierarchical architectures and has been successfully applied in many areas such as image processing, speech recognition and natural language processing. This talk will discuss the use of deep-learning in intrusion detection systems and outline some of the applications that we have been recently working on.
In this presentation, the authors propose latent representation models for improving network anomaly detection. Well-known anomaly detection algorithms often suffer from challenges posed by network data, such as high dimension and sparsity, and a lack of anomaly data for training, model selection, and hyperparameter tuning. Our approach is to introduce new regularizers to a classical Autoencoder (AE) and a Variational Autoencoder (VAE), which force normal data into a very tight area centered at the origin in the non-saturating area of the bottleneck unit activations. These trained AEs on normal data will push normal points towards the origin, whereas anomalies, which differ from normal data, will be put far away from the normal region. The models are very different from common regularized AEs, Sparse AE and Contractive AE, in which the regularized AEs tend to make their latent representation less sensitive to changes of the input data. The bottleneck feature space is now used as a new data representation. A number of one-class learning algorithms are used for evaluating the proposed models. The experiments testify that our models help these classifiers to perform efficiently and consistently on high-dimensional and sparse network datasets, even with relatively few training points. More importantly, the models can minimize the effect of model selection on these classifiers since their performance is insensitive to a wide range of hyperparameter settings.
Cybersecurity education and training activities are critical for increasing the resilience of government institutions, organization and companies when faced with the cyberattacks that are occurring on an ever-increasing scale. In this talk, an integrated cybersecurity training framework, named CyTrONE, will be introduced. CyTrONE is designed and implemented to address the above shortcomings by automating the training content generation and environment setup tasks. This talk will discuss the architecture and implementation of the framework, then also present its evaluation from functionality and performance perspectives.
Intrusion detection based upon machine learning is currently receiving a considerable interest from the research community. One of the appealing properties of machine learning based intrusion detection system is their ability to detect new or unknown attacks. When applying machine learning to intrusion detection, it is required that large number of both attack and normal data samples are collected. While, it is often easier to sample benign data based on the normal behavior of networks, intrusive data is much more scarce and difficult to collect. In this research, we propose a novel solution to this problem by using generative adversarial networks to generate synthesized attack data for intrusion detection. The synthesized attacks are merged with the original data to form the augmented training dataset. Three popular machine learning techniques are then trained on the augmented dataset. The experiments conducted on two common intrusion detection datasets that show that machine learning algorithms achieve better performance when trained on the augmented dataset of the generative adversarial networks compared to those trained on the original dataset and other sampling techniques.
Beside many other algorithms such as Singular Value Decomposition (SVD), Discreet Cosine Transform (DCT) and Discrete Wavelet Transform (DWT), QR decomposition is known as an effective method for embedding and extracting watermarked image. In QR decomposition, Gram-Schmidt and Householder are the most popular two algorithms. In this paper, QR factorization is executed by exploiting orthogonality of Q matrix and triangularity of R matrix to find out elements of these matrixes. The algorithm of Sun is used for both embedding and extracting. For embedding watermark, Q and R will be calculated separately where computing R is implemented by solving a set of linear equations and calculating Q is base on Gram-Schmidt algorithm via knowing R. In addition, diagonal elements of R matrix are inspected to ensure their validity except the first diagonal one. For extracting watermark, only the first element R(1,1) of R matrix needs to be computed which is performed by an operation. Experimental results show that the proposed scheme not only has better quality of watermarked image, but also overcomes problems of robustness and computation complexity.
For detail please contact Dr. Trangdldtrang@gmail.com +84 971001199
Phòng 306 nhà H2
Số 236, Hoàng Quốc Việt, Cầu Giấy, Hà Nội