Project Start Date:
September 2016
Project Status:
On-going
Project Objectives and Scope
We research and design algorithms, technologies and systems of big data security analytics. We research effective algorithms and system designs that enable efficient analysis of big data repositories, uncover hidden relationships within massive amounts of security data and identify advanced threats and cyber-attacks in their early stages.
Challenges
The main challenge is to be able to efficiently perform long-term analytics on large-scale and massive volume of heterogeneous security data, detect the attack at an early stage, detect unseen attack and zero-day exploit, understand how the initial penetration to the organization occurred, and derive the goal of the attacker and estimate the damage if the attack is running for an extended period of time.
Methodology/Approach
In this project we research novel detection algorithms based on data-driven approach, supervised, semi-supervised and unsupervised learning approaches, and state-of-the-art deep learning approaches. We research new time series machine learning algorithms that can model the system over time. Time series machine learning algorithms, in particular, Recurrent Neural Networks, have the ability to model what they have seen in the past. This concept would be useful for the detection of APTs as these attacks have low profiles with long-term execution.
Publications
On data augmentation for GAN training
InfoMax-GAN: Improved Adversarial Image Generation via Information Maximization and Contrastive Learning
Attentive Weights Generation for Few Shot Learning via Information Maximization
Attention-based Context Aware Reasoning for Situation Recognition
Self-supervised GAN: Analysis and Improvement with Multi-class Minimax Game
A Neural Attention Model for Real-Time Network Intrusion Detection
Leveraging Multi-aspect Time-related Influence in Location Recommendation
An Improved Self-supervised GAN via Adversarial Training
Few-Shot Regression via Learned Basis Functions
Improving GAN with Neighbours Embedding and Gradient Matching
DOPING: Generative Data Augmentation for Unsupervised Anomaly Detection with GAN
Dist-GAN: An Improved GAN using Distance Constraints
Mining Subgraphs From Propagation Networks Through Temporal Dynamic Analysis
Efficient and Deep Person Re-Identification using Multi-Level Similarity
Exploiting Reshaping Subgraphs From Bilateral Propagation Graphs
Analyst Intuition Inspired Neural Network Based Cyber Security Anomaly Detection
Adaptive Quantization for Deep Neural Network
Recurrent Neural Network (RNN) Base Computational Model for Cyber Attack Detection
Analyst Intuition Based Hidden Markov Model on High Speed, Temporal Cyber Security Big Data
Analyst intuition inspired high velocity big data analysis using PCA ranked fuzzy k-means clustering with multi-layer perceptron (MLP) to obviate cyber security risk
Principal Investigator(s)

CHEUNG, Ngai-Man (Man)
- ngaiman_cheung@sutd.edu.sg
- +65 6499 4542
- 1.502.17
- SUTD Profile
-
- Big Data Security Analytics (Principal Investigator)
- Predicting Adversarial Behaviours and the Motivation for Automated Network Defense (Co-Principal Investigator)
- Trusted and Resilient Monitoring Infrastructure (Co-Principal Investigator)

TAN, Chee Hiong
Co-Principal Investigator(s)

LU, Wei
- wei_lu@sutd.edu.sg
- +65 6499 4784
- 1.302.10
- SUTD Profile
-
- Big Data Security Analytics (Co-Principal Investigator)
Researcher(s)

TRAN, Ngoc Trung
- ngoctrung_tran@sutd.edu.sg
-
- Big Data Security Analytics (Researcher)




TRAN, Viet Hung
- viethung_tran@sutd.edu.sg
-
- Big Data Security Analytics (Researcher)