NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA FACULTY BRIEFINGS Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Wireless Communications and Networking M. CENK GURSOY Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Wireless World • There were 4.3 billion global mobile users in 2014. 5.2 billion users are predicted by 2019. • Global mobile data grew 69% in 2014 and reached 2.5 exabytes (1018) per month. • Almost half a billion (497 million) mobile devices and connections were added in 2014. • It is predicted that there will be 11 billion mobileconnected devices by 2019. • Mobile video traffic grew to 55% by the end of 2014 and will represent 72% of the total mobile data traffic by 2019. 2 Wireless Challenges •Wireless medium is a timevarying, open, shared, and broadcast medium. •Resources, such as energy and bandwidth, are limited and have to be efficiently used. •Exponential growth in mobile data and multimedia content further strains the limited resources. •We are experiencing a spectrum crunch, and carbon footprint of wireless services is growing. 3 1. Green Wireless Communications •Due to limited energy available for mobile units, rising energy costs, and environmental concerns, energy efficiency in wireless systems is a critical concern. •We investigate –Energy-efficient wireless transmission schemes, –Energy-efficient resource allocation strategies,, –Energy-efficient power control policies, –Energy-efficient networking protocols, –Wireless power transfer, RF energy harvesting, and wireless-powered networks. 4 2. Cognitive Radio and mmWave Systems •Prime portion of the radio spectrum (30 MHz – 3 GHz) has already been allocated to specific services and applications. •We study two solutions to the spectrum scarcity problem: –Cognitive radio systems to overcome spectrum underutilization •Channel sensing errors, interference constraints, throughput, error rates. –Moving to the less explored, less utilized millimeter wave band (30 – 300 GHz) 5 3. Multimedia over Wireless Networks •Multimedia traffic (e.g., VoIP, video streaming, interactive video, online gaming) requires certain quality of service guarantees. –We take into account the randomness of the multimedia traffic and investigate the throughput under statistical delay/buffer constraints. –We design multimedia transmissionreception systems and analyze the quality of received images and videos. –We determine efficient resource allocation strategies to maximize the performance and multimedia quality. 6 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Fighting Android Malware Heng Yin Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center 1 Android Malware: An Increasing Threat McAfee Threat Report: Totaled 3.73 million samples at the end of 2013, a 197% increase over 2012 2 Malware Variants and Zero-day Malware McAfee Threat Report: 2.47 million new mobile malware samples were collected in 2013 3 My Research Samples • DroidScope [USENIX Security 2012] – For security analysts – Understand malware’s innerworkings • DroidSIFT [ACM CCS 2014] – For app markets & antivirus scanners – Detect zero-day malware and new variants • DescribeMe [under submission] – For end users – Generate security-centric description – Deter installation of malware 4 DroidSIFT • Two key aspects – Extract Weighted Contextual API Dependency Graphs as program semantics – Introduce Graph Similarity metrics to address malware variants and zero-day malware 6 Evaluation: Classification Results • Detection of Transformation Attacks 7 DescribeMe: Automatic Generation of SecurityCentric Description When downloading an app, a user sees the following: 8 This is our description The app retrieves your phone number and writes to output depending on if the user selects Button and prepares to write data to network. The app retrieves text from user input and writes to output depending on if the user selects button and prepares to write data to network. 9 Impacts of New Descriptions App Download Rates Old Description New Description Malware 63.4% 24.7% App /w Privacy Leak 80% 28.2% Clean Apps 71.1% 59.3% 10 Questions? http://lcs.syr.edu/faculty/yin 11 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Jae C. Oh Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Distributed Multi-Agent Laboratory • Big Data: Analysis, Computations, and Visualizations – Recommender Systems and Reputation Management – Visualizations – Resource management in Cloud computing • Massively Multi-agent and Multi-Robot Problems – Autonomous Agents and Robots For Search and Rescue – Distributed Task Allocation – Game Theoretical Cyber-Physical Systems 2 Reasoning About Interactions and Relationships • Reasoning/predicting about what each rational agent would do in a particular situation (e.g., decision support) • Anti-terrorism, rapid reallocation of resources. • Finding robust communication, resource routing in a damaged infrastructure. • Credit Card Fraud Detection through cost/benefit analysis • Spam filtering, identifying important document/information. 3 Big Data Fusion and Visualizations • • • • • Analysts working the information security space are confronted with a number of challenges in their efforts to collect, process, and analyze data in order to produce timely, accurate, and meaningful intelligence for the protection of computing infrastructure and systems. The volume and diversity of both open source and proprietary data available make it increasingly difficult for the analyst to collect and process. The intent of the project is to produce a prototype fusion and visualization system. A prospective system would be capable of aggregating federated structured and unstructured data. The system would perform multi-level fusion wherein raw data with differing origin, format, content, and context receive the appropriate level of processing for effective relationship and predicative analytics. Fused data would then be used to synthesize intelligence which would be represented in a number of manners including visualization and alerts The anticipated prototype will provide two new capabilities to the Cyber Threat and Security Operation Analysts. The first is derived from the multi-level fusion wherein key pieces of data gleaned from highly structured network telemetry is enriched with unstructured data and semantic meaning and used as measurement for acceptable, unacceptable, and anonymous communication and behavior of networked systems. The second is the representation of fused data to provide the analyst more flexible and intuitive approaches to comprehend and correlate activities while investigating incidents of interest. The team will research solutions to challenges in the collection and processing of data necessary to detect information security incidents and fraudulent online banking transactions. The outcome will be a prototype system which JPMC’s analysts may use to comprehend complex data, investigate, and improve downstream alerting on issues of interest. Most organizations have limited capability to identify and extract key attributes and leverage them for analysis as well as guiding further collection. The analysis and reporting processes are impacted by limited options for representing and visualizing data resulting in slower or incomplete analysis and comprehension of impact. 4 Massively Distributed Multi-Robots • Goal: Design software and hardware that work in the real physical world. Example: Search and Rescue 5 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA G-Storm: A GPU-accelerated Platform for Online Big Stream Data Processing Jian Tang Electrical Engineering and Computer Science Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center A General Platform for Online Big Stream Data Processing (Storm/S4/TimeStream/MillWheel) 2 Challenges GPU for Online Stream Data Processing, really? • Different programming models • Different data processing models • Can GPU’s new features be utilized? 3 GPU-accelerated Big Stream Data Processing 4 G-Storm: Desirable Features G-Storm: the first GPU-accelerated general-purpose parallel platform for online big stream data processing. • • • • A general (rather than application-specific) data processing platform as Storm, which can deal with various applications and data types. Exposing GPUs to Storm applications while preserving its easy-to-use programming model by handling parallelization, data transfer and resource allocation automatically via a runtime system without user involvement. Achieving high-throughput and low-overhead data processing by buffering multiple data tuples and offloading them in a batch to GPUs for highly-parallel processing. Accelerating data processing further by supporting Direct Data Transfer (DDT) between two executors co-located on a common GPU. 5 G-Storm: Software Architecture 6 G-Storm: Experimental Results with the Continuous Query App 7 G-Storm: Experimental Results with the Continuous Query App 8 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Kevin Du Cybersecurity FACULTY BRIEFINGS Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center The Summary of My Work Attack Defense Research Education Attacks on Smartphone • Objectives: – Systematically study the design of Android system – Find attacks – Understand Fundamental causes – Solution and Automatic detection Code Injection Attack • Discoveries: – WebView (2011, led to a grant from Google) – Code Injection (2014) – Multi-user Framework (2014) – Clipboard (2014) – Data Residue (2015) – Dangerous Assumption (2015) News Coverage Re-design Android’s Security • Objectives: Protect Location – Systematically study the weakness of the security design in Android – Develop innovative solutions to improve the design – Get them adopted by Android • Outcomes – Isolate ads from apps – SEIntent Firewall (undergrad) – Resource isolation Input Method Security Education • Objective – Develop lab exercises • Outcome – – – – – – – – 13 years:since 2002 $1.3 million: 3 NSF grants 30: number of labs developed 700 pages of lab materials $0: cost of using these labs 250 schools 30 countries 70,000 hits per month • This Summer – 60 instructors come to SU for training NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Kishan Mehrotra Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Why Big Data Analysis? More data -> More information Competitive advantage Hidden patterns, unknown correlations Effective marketing, customer satisfaction, increased revenue Better business decisions Dynamic nature of data 2 Key Player problem: Large Social Networks • Existing single objective key player identification algorithms – Eigen-value Centrality, Closeness Centrality, … • Multi objective algorithm out performs traditional single objective algorithms – time complexity is large. • Degree based network sampling was proposed to reduce time complexity. • Facebook (4039, 88234) – {(17 min, 100%), (13 min, 70%), (8 min, 30%)} Classification for Big data -- Binary Classifier based on Conjunction Rule • Naïve Bayes for Binary Data • B1 ˄ B2 ˄ ┐B3 ˄ … ˄ BM => (C = 1) • Contingency Table is used as the sufficient statistic Attribute Class Total 0 1 0 N00(j) N01(j) No 1 N10(j) N11(j) N1 Performance Improvement • Compared to CART[2] • Faster by a factor of log N • For M = 500 and N = 30,000 • Time for CART = 3 minutes • Time for BCC = 0.038 seconds • Proposed algorithm is suitable for Data Streams ---Just update the contingency tables [2] Denison, David GT, Bani K. Mallick, and Adrian FM Smith. "A bayesian CART algorithm." Biometrika 85.2 (1998): 363-377. 5 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Melissa A. Green Mechanical and Aerospace Engineering Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Coherent structures in fluid dynamics Coherent structures in fluid dynamics • Bio-inspired propulsion Coherent structures in fluid dynamics • Bio-inspired propulsion • Vortex-induced unsteady forces Coherent structures in fluid dynamics • Bio-inspired propulsion • Vortex-induced unsteady forces • Turbulence Coherent structures in fluid dynamics • Bio-inspired propulsion • Vortex-induced unsteady forces • Turbulence • Turbulence + combustion Coherent structures in fluid dynamics • Bio-inspired propulsion • Vortex-induced unsteady forces • Turbulence • Turbulence + combustion Lagrangian coherent structures • Coherent motions or patterns in complex 3D unsteady fluids • Ridges of finite-time Lyapunov exponent yield dynamically relevant manifolds Lagrangian coherent structures • Coherent motions or patterns in complex 3D unsteady fluids • Ridges of finite-time Lyapunov exponent yield dynamically relevant manifolds Lagrangian coherent structures • Coherent motions or patterns in complex 3D unsteady fluids • Ridges of finite-time Lyapunov exponent yield dynamically relevant manifolds Lagrangian coherent structures • Coherent motions or patterns in complex 3D unsteady fluids • Ridges of finite-time Lyapunov exponent yield dynamically relevant manifolds • Intersections are critical points Lagrangian coherent structures • Coherent motions or patterns in complex 3D unsteady fluids • Ridges of finite-time Lyapunov exponent yield dynamically relevant manifolds • Intersections are critical points • Low-dimensional way to track dynamically-relevant structures in the complex field Lagrangian coherent structures • Coherent motions or patterns in complex 3D unsteady fluids • Ridges of finite-time Lyapunov exponent yield dynamically relevant manifolds y/h • Intersections are critical points • Low-dimensional way to track dynamically-relevant structures in the complex field U • Potential in fluid dynamics for flow control, drag reduction, efficient propulsion • Tools could be useful in problems where dynamics can be captured from differential equations or trajectory tracking NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Embedded Computer Vision and Mobile Smart Camera Applications Monday, APRIL 6, 2015 Senem Velipasalar Electrical Engineering and Computer Science Department Smart Vision Systems Laboratory vision.syr.edu Sheraton Syracuse University Hotel & Conference Center Wireless Embedded Smart Cameras A battery-powered, wireless embedded smart camera is a stand-alone unit that not only captures images, but also includes a processor, memory and communication interface, and thus combines sensing, processing and communication on a single embedded platform. Application areas for smart cameras include cyber-physical systems, video surveillance systems, intelligent transportation systems, UAVs, medical applications, elder care, robotics, wildlife monitoring, entertainment, manufacturing and inspection. Battery-powered embedded smart cameras introduce many additional challenges including: Limited energy - 4AA batteries Limited processing power and memory Limited bandwidth Smart Vision Systems Laboratory Syracuse University This work has been funded in part by NSF CAREER grant CNS1206291 and NSF Grant CNS-1302559. 2 Indoor Scene Understanding for Vision Based Smart Autonomous UAVs Towards smart and autonomous LBP- Based Face Tracking: Object Detection and Navigation: Face Tracking Video Demo: UAVs for indoor applications Java + OpenCV based frame work for computer vision applications Smart Vision Systems Laboratory Syracuse University This work has been funded in part by NSF CAREER grant CNS1206291 and NSF Grant CNS-1302559. 3 Wearable Camera- and Accelerometer-based Fall Detection on Portable Devices We developed a fall detection system using wearable devices, e.g. smart phones and tablets, equipped with cameras and accelerometers. Since the portable device is worn by the subject, monitoring is not limited to confined areas, as opposed to static sensors installed at certain rooms. Fusing camera and accelerometer data not only increases the detection rate, but also significantly decreases the number of false alarms. Experimental results and trials with actual Samsung Galaxy phones show that combining two different sensor modalities, provides much higher sensitivity, and a significant decrease in the number of false positives during daily activities. Android smartphone attached to belt Smart Vision Systems Laboratory Syracuse University This work has been funded in part by NSF CAREER grant CNS1206291 and NSF Grant CNS-1302559. 4 Energy-efficient Depth-Image Assisted Object Detection on Smart Phones Object detection is computationally demanding, and processing power and energy are limited resources for mobile devices. We efficiently perform object detection by incorporating depth image, and integrating dynamic voltage frequency scaling (DVFS). The search region is narrowed down by applying a threshold on depth image. A ground removal algorithm is applied to further reduce the search regions. Haar feature-based object detection is performed in the reduced area of RGB image. If an object of interest cannot be found, the frequency of the processor is scaled down. If an object of interest is detected, the frequency of the processor is increased. Average energy savings can reach up to 38.90%. Smart Vision Systems Laboratory Syracuse University This work has been funded in part by NSF CAREER grant CNS1206291 and NSF Grant CNS-1302559. 5 Other Research Activities Tracking of Vehicle Taillights and Alert Signal Detection by Embedded Smart Cameras We developed a robust and computationally lightweight algorithm for a real-time vision system, capable of detecting and tracking vehicle taillights, recognizing common alert signals using a vehicle-mounted embedded smart camera, and counting the cars passing on both sides of the vehicle. The system processes scenes entirely on the microprocessor of an embedded smart camera. Mobile Standards-Based Traffic Light Detection in Assistive Devices for Individuals with CVD A robust, traffic-standards-based, and computationally efficient method for detecting the status of the traffic lights without relying on Global Positioning System, lidar, radar information, or prior (mapbased) knowledge. The system can accurately identify the status of the light at 400 ft away from the intersection, reliably detecting solid, faulty, arrow, and high-visibility signal lights. Over 50 h of video (over 2000 intersections) were tested with the system with 97.5% accuracy of solid light detection. Scale Estimation with Difference of Ordered Residuals We developed a scale estimator and applied it to multiple model estimation problems for detecting planes and two-view motions. Smart Vision Systems Laboratory Syracuse University This work has been funded in part by NSF CAREER grant CNS1206291 and NSF Grant CNS-1302559. 6 vision.syr.edu http://eng-cs.syr.edu/faculty/velipasalar/ Smart Vision Systems Laboratory Syracuse University This work has been funded in part by NSF CAREER grant CNS1206291 and NSF Grant CNS-1302559. NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Nangia Research Group Department of Biomedical and Chemical Engineering Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Computational modeling of biological systems Designing anticancer nanocarriers Quantum dots-protein corona Bacterial membranes Antimicrobial peptides Blood brain barrier tight junctions Cellular Uptake of nanoparticles Protein structure of Ebola virus 2 Length and Time scale challenge 3 Current Solutions • High Performance Computing resources. Codes are tightly coupled MPI & OpenMP • Visualization –Images, animations, trajectories that better represent the phenomena • Storage –Archiving volumes of data that is readily available for retrieval STAMPEDE CLUSTER 4 “Bigger data” ahead…. Variety Particle-Positions/Momenta/ Accelerations/Forces/Pressure Velocity Femtosecond/Picosecond/Nano second/microsecond/ millisecond/second/hour/days Volume 7× 1027 atoms Megabytes/gigabytes/terabytes /petabytes/zetabytes 5 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Vir V Phoha Department of Electrical Engineering & Computer Science FACULTY BRIEFINGS Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Attack Defense Research Area Research Manipulation of Learning Systems– Medical Cyber Physical Systems Foundation Research and algorithms Breaching authentication Video analytics; Robotic attacks Foundation Research and algorithms Prototype Mobile biometric authentication Foundation Research and algorithms Prototype DARPA Prototype DARPA Desktop biometric authentication Technology Funding Source White paper accepted: Invited Proposal by DHS 2 Adaptive systems that learn from examples Attacks on Learning (adaptive) systems Adaptive Implantables • Rate adaptive pacemakers • Implantable pulse generators Adaptive control in cars Web based systems • Spam filters • Intrusion Detection Systems Mobile behavioral biometric authentication 3 Reasons that an attack will work Variation in the signatures Temporal drift Overlap of patterns over large populations sets 4 Defense --DARPA TA1a and TA1b Active Authentication TA1a: Desktop Atomic Keystroke Features (Enhanced) Higher Level Keystroke Features TA1b: Mobile Sensor Readings Typing (T) Swiping (S) Cognilinguistic Demographic Accelerometer (A) T+A A S+A A S+A Behavior Modeling and Fusion Authentication, Feedback for Template Update © Louisiana Tech University Experiment Results (Desktop) Atomic Keystrokes: Accuracy 96.641%, FAR .0295, FRR 0.0382 Experiment: (Desktop) 831 volunteer participants (Mobile) 178 volunteers participants The above results were obtained using typing data collected from volunteers in a single application context. Preliminary work indicates that system accuracy with passively collected data from multi-application, multi-window environments may initially be lower. (Mobile) Swiping: FAR: 10.3%, FRR: 10.3% Typing: FAR: 19%, FRR: 19% Body Movements (Gait): FAR: 7.9% , FRR: 7.9% Typing + Accelerometer : FAR: 13.2%, FRR: 13.2% 5 NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Yingbin Liang Department of Electrical Engineering & Computer Science Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Machine Learning of Big Data Challenges: Ultra high dimension: Machine Learning data size grows rapidly(GB, TB), up to billion features. Existing ML algorithms does not scale well or even fail Our Solutions: Muti-task learning Truncating large problem Distributed and parallel implementation Study 1: Structure Learning of Large Networks: Simultaneously learn multiple networks Sample size is significantly reduced Study 2: Feature Selection in Human Genome Truncate feature set Sequential and parallel processing Theoretical performance guarantee Asynchronous ML Algorithm Fully synchronized distributed systems have communication bottleneck Asynchronous System Avoids communication bottleneck More efforts can be spent on computational work i = 1,..., k c +1 i x k machines c = x + ∑ U i ( xit ) local fresh updates 0 i k +∑ j =1\ i t =0 Tic, j t U x ( ∑ j j ) delayed updates t =0 c 0 ≤ c − Ti , j ≤ s Strong Theoretical Gurantees for 2 l0 , l1 , groupl0 , groupl1 , x 2 regularized non-convex Lasso Type problems: Algorithm converges to critical point with finite length ∞ ∑ from other machines bounded delay O(1/c) x c +1 − x c < +∞ c =0 rate in convex case ( ) ( ) F x c +1 − F x c ≤ O(1 / c ) Detection of Anomalous Patterns Anomalous segment Existing approaches detects only based on individual quantities Our problem: Anomalous patterns x x x x x x x x distribution space x x x x x x Sample sets from different distributions o o o oo o o o o o o o kernel o p q Hilbert space Our approach: kernelbased method MMD[p,q] Nonparametric nature: no prior information about patterns is required Guaranteed performance for arbitrary patterns Dirty Interference Cancellation for 4G and LTE systems Wireless Networks Current Interference management is an critical issue Using orthogonal resources Cellular User D2D Transmitter 2 Message D2D Receiver 2 Our new perspective Base Station Simultaneous transmission Interferer deals with interference it causes Cancel large power interference with small additional power PHY-Based Secret Sharing and Secret Key Generation Traditional Approach Number theory based technique for secret key sharing Trusted third party Public Discussion Eve Our Apporach SK PK Exploit physical channel randomness Flexible to be generated to multisecret sharing and multi-key generation Characterization of performance limits over all possible approaches NUNAN RESEARCH DAY CYBER, WIRELESS and BIG DATA Yuzhe Tang Department of Electrical Engineering & Computer Science Monday, APRIL 6, 2015 Sheraton Syracuse University Hotel & Conference Center Cloud computing and big data The era of cloud computing has arrived: Daily life changed by cloud computing Many cloud service providers in the market Human activities Cloud service providers 2 Cloud Computing Barrier • One hand, people reluctant to using Cloud. –Security/lack of trust are top concerns. 3 Cloud Computing Barrier • One hand, people reluctant to using Cloud. –Security/lack of trust are top concerns. •The other hand, cloud not trustworthy –Caught being “evil” all the time Cloud service provider PRISM scandal! 4 My Research: Overcome the Barrier My research is to bridge the gap by building Secure & Trustworthy Distributed Systems Two current focuses: 1. Trustworthy key-value stores 2. Efficient multi-party computations for federated data analysis 5 Research Details Trustworthy Key-Value Stores -Verifiable data freshness & consistency -NewSQL with predictable performance Efficient Multi-Party Computations -Scalable MPC -Tunable MPC on heterogeneous hardware 6 Thank you. Contact: Yuzhe Tang Assistant Professor Syracuse University [email protected] http://ecs.syr.edu/faculty/yuzhe
© Copyright 2024