CyPSEC Lab Research

Generative Adversarial Networks for Security and Privacy

One of the major thrusts in CyPSEC lab is on Generative Adversarial Networks. Generative adversarial network (GAN) has attracted great attention due to its impressive performance and potential numerous applications, such as data augmentation, real-like image synthesis, image compression improvement, etc. The generator in GAN learns the density of the distribution from real data in order to generate high fidelity fake samples from latent space and deceive the discriminator. GANs have been majorly used in imaging applications. In CyPSEC lab we explore the functionalities of GANs in and beyond imaging applications. CyPSEC lab researchers focus majorly on data augmentation in network intrusion detection settings, privacy leakages and protection while training with sensitive data samples, various input+model+output perturbation mechanisms to guarantee privacy and build resilience against model inversion and known attacks.

Communication Security and Privacy in Smart Homes, Smart Grids and SCADA Environments

The technical advances, especially in Internet of Things (IoT) helps in forming enhanced monitoring and control environments in smart homes, grids especially in SCADA settings through connectivity between devices. This communication being greatly reliant on information technology, concerns regarding security and privacy aspects inevitably creep in. CyPSEC lab broadly focuses on communication protocol security and privacy issues that arise in smart homes settings, smart home-smart grid interactions and SCADA environments. This focus area identifies the threats to individual's privacy in smart homes that adjust their energy profile to the dynamic real time prices (RTP)/time of use (TOU) tariffs to minimize their expenditure, and report attributable sensitive fine-grained data to maintain demand in line with supply during peak hours of usage. On the grids and SCADA side, we particularly focus on security in Modbus, DNP3 and GOOSE protocols; and focus on testing and integration of cryptographic features to the stack and assess the feasibility of practical implementation to enhance grid protection.

Privacy Preserving Data Analytics - Differential Privacy and Computation over Encrypted Data

This research thrust in CyPSEC lab focuses on enabling practical, accurate and differentially private data analyses on large datasets, and on the tradeoff between the achieved privacy and accuracy under tunable privacy settings. We also focus on optimization issues that arise with data scaling, correctness in data analysis, and information leakage measurements. Analysis of large datasets of potentially sensitive private information about individuals raises natural privacy concerns. Existing privacy preserving techniques like, anonymization requires having dataset divided in the set of attributes like sensitive attributes, quasi identifiers, and non-sensitive attributes. Differential privacy is a recent area of research that brings mathematical rigor to the problem of privacy-preserving analysis of data. Informally the definition stipulates that any individual has a very small influence on the (distribution of the) outcome of the computation. Thus an attacker cannot learn anything about an individual's report to the database, even in the presence of any auxiliary information he/she may have. A large and increasing number of statistical analyses can be done in a differentially private manner while adding little noise. We make this possible by exploring deep connections between learning theory, convex geometry, communication complexity, cryptography and robust statistics.

Network Economics: Strategy-proof Allocations for Spectrum and Demand Response/Management

In CyPSEC lab we focus on strategy proof allocation frameworks for spectrum trading and demand side management to provide incentives. Most of these architectures are designed based on Matching Theory (Nobel-prize winning microeconomic framework), various auction mechanisms and pricing mechanisms. These architectures consider threats from external adversaries as well as internal attackers (semi honest/colluding auctioneers, honest but curious operators etc.) in design to create strategy proof mechanisms that are practical as well as guarantee truthfulness in allocations. In spectrum settings, we jointly integrate these strategy proof frameworks with spectrum reusability. In case of demand side management we integrate these with optimization in emergency demand response scenarios for data centers and smart grids to incentivize energy usage reduction.

Intelligent and Safe Transportation [Intelligent Route Scheduling, In-vehicle Security]

One of the research thrusts of CyPSEC lab is on security and privacy in Vehicular Networks. Some of the areas we primarily focus are on intelligent route scheduling for transportation network company (TNC) services and on security in in-vehicular networks such as CAN bus security. Some of the questions we address are on reducing the cruising time while improve the efficiency/earnings through deep reinforcement learning-based TNC route scheduling approach. We consider multiple factors in the complex and dynamic TNC environment, such as locations of the TNC vehicles, different time periods during the day, the competition among TNC vehicles, etc., to design deep Q-network-based route scheduling algorithms operating under distributed framework, which moves the server closer to the terminal users and accelerates the training speed. Furthermore, we protect sensitive location information uploaded by the passengers through geo-indistinguishability scheme based on differential privacy and evaluate performance in such use-cases using open real-time data sets. Along the lines of in-vehicular network security we investigate vulnerabilities in CAN bus that connect ECUs in cars to develop deterrent and corrective controls.

Google Sites

Report abuse