Defense Notices

Neural Networks were studied and applied to many research problems from a long time. With gaining popularity of deep neural networks in the area of machine learning, many researchers in various domains want to try deep learning framework. Deep learning requires lot of memory and high processing power. One way of doing it faster is to make use of GPUs which use distributed and parallel processing, thereby increasing speed. But because of the computation (lot of vector and matrix operations) deep learning requires, expensive infrastructure required (GPUs and clusters), hardware and software installation overhead, not many researchers prefer deep learning. The current application is a solution to cheminformatics problems using Convolutional Architecture for Fast Feature Embedding (Caffe) deep learning framework. The application provides a framework/service to researchers who want to try deep learning on their datasets. The application accepts datasets from users along with options for hyper parameter configuration, runs cross fold validation on the training dataset, and makes predictions on the test dataset. The (tuning) results of running caffe on the training dataset and predictions made on test dataset are sent to user via an email. The current version supports binary classification that predicts activity/inactivity of a chemical compound based on molecular fingerprints which are binary features.

Given the high complexity and increasing traffic load of the current Internet, the geographically-correlated challenge caused by large-scale disasters or malicious attacks pose a significant threat to dependable network communications. To understand its characteristics, we start our research by first proposing a critical-region identification mechanism. Furthermore, the identified regions are incorporated into a new graph resilience metric, compensated Total Geographical Graph Diversity (cTGGD), which is capable of characterizing and differentiating resiliency levels for different topologies. We further propose the path geodiverse problem (PGD) that requires the calculation of a number of geographically disjoint paths, and two heuristics with less complexity compared to the optimal algorithm. We present two flow-diverse multi-commodity flow problems, a linear minimum-cost and a nonlinear delay-skew optimization problem to study the tradeoff among cost, end-to-end delay, and traffic skew on different geodiverse paths. We further prototype and integrate the solution from above models into our cross-layer resilient protocol stack, ResTP--GeoDivRP. Our protocol stack is implemented in the network simulator ns-3 and emulated in the KanREN testbed. By providing multiple geodiverse paths, our protocol stack provides better path protection than Multipath TCP (MPTCP) against geographically-correlated challenges. Finally, we analyze the mechanism attackers could utilize to maximize the attack impact and demonstrate the effectiveness of a network restoration plan. 

In a world with enormously growing technologies and applications, most people use smart 
devices. This provides a means to develop smart applications that will be help students learn effectively. 
In this project, we develop a smart android application which will provide digital means of 
interaction between the professors and students. Instead of using traditional emails for every 
discussion, this application helps to broadcast multiple messages to the class through a single 
click. The students will also be able to follow multiple professors and participate in the active 
discussions. And also this application allows the users to send personal messages to the other 
users in order to participate in an active discussion. It provides unique logins to every student 
and professor. It uses mongoDB as the database and "parse" backend as a service.The main 
inspiration for this project was an application called Tophat. 

As the passenger vehicles evolve to be “smart”, electronic components, including communication, intelligent control and entertainment, are continuously introduced to new models and concept vehicles. The new paradigm introduces new features and benefits, but also brings new security issues, which is often overlooked in the industry as well as in the research community. 

Smart cars are considered cyber-physical systems (CPS) because of their integration of cyber- and physical- components. In recent years, various threats, vulnerabilities, and attacks have been discovered from different models of smart cars. In the worst- case scenario, external attackers may remotely obtain full control of the vehicle by exploiting an existing vulnerability. 

In this research, we investigate smart cars’ security from a CPS’ perspective and derive a taxonomy of threats, vulnerabilities, attacks, and controls. In addition, we investigate three security solutions that would improve the security posture of automotive networks. First, as automotive networks are highly vulnerable to Denial of Service (DoS) attacks, we investigate a solution that effectively mitigates such attacks, namely ID-Hopping. In addition, because several attacks have successfully exploited the poor separation between critical and non-critical components in the automotive networks, we propose to investigate the effectiveness of firewalls and Intrusion Detection Systems (IDS) to prevent and detect such exploitations. To evaluate our proposals, we built a test bench that is composed of five microcontrollers and a communication bus to simulate an automotive network. Simulations and experiments performed with the testbed demonstrates the effectiveness of ID-hopping against DoS attacks. 

One source of insight into the motivations of a modern human being is the text they write and post for public consumption online, in forms such as personal status updates, product reviews, or forum discussions. The task of inferring traits about an author based on their writing is often called "author profiling." One challenging aspect of author profiling in today’s world is the increasing diversity of natural languages represented on social media websites. Furthermore, the informal nature of such writing often inspires modifications to standard spelling and grammatical structure which are highly language-specific. 
These are some of the dilemmas that inspired a series of so-called "shared task" competitions, in which many participants work to solve a single problem in different ways, in order to compare their methods and results. This thesis describes our submission to one author profiling shared task in which 22 teams implemented software to predict the age, gender, and certain personality traits of Twitter users based on the content of their posts to the website. We will also analyze the performance and implementation of our system compared to those of other teams, all of which were described in open-access reports. 
The competition organizers provided a labeled training dataset of tweets in English, Spanish, Dutch, and Italian, and evaluated the submitted software on a similar but hidden dataset. Our approach is based on applying a topic modeling algorithm to an auxiliary, unlabeled but larger collection of tweets we collected in each language, and representing tweets from the competition dataset in terms of a vector of 100 topics. We then trained a random forest classifier based on the labeled training dataset to predict the age, gender and personality traits for authors of tweets in the test set. Our software ranked in the top half of participants in English and Italian, and the top third in Dutch.

As one of the world’s most famous online social networks (OSN), Twitter now has 320 million monthly active users. Accompanying the large user group and abundant personal information, users increasingly realize the vulnerability of tweets and have reservations of showing certain tweets to different follower groups, such as colleagues, friends and other followers. However, Twitter does not offer enough privacy protection or access control functions. Users can just set an account as protected, which results in only the user’s followers seeing the tweet. The protected tweet does not appear in the public domain, third party sites and search engines cannot access the tweet. However, a protected account cannot distinguish between different follower groups or users who use multiple accounts. To serve the demand of the user so that they can restrict the access of each tweet to certain follower groups, we propose a browser plug-in system, which utilizes CP-ABE (Ciphertext Policy Attribute based encryption), allowing the user to select followers based on predefined attributes. Through simple installation and pre-setting, the user can encrypt and decrypt tweets conveniently and can avoid the fear of information leakage.

The shared resources in multi-core systems, mainly the memory subsystem(caches and DRAM), if not managed properly would affect the predictability of real-time tasks in the presence of co-runners. In this work, we first studied the design of COTS DRAM controllers and its impact on predictability and, proposed a DRAM controller design, called MEDUSA, to provide predictable memory performance in multi-core based real-time systems. In our approach, the OS partially partitions DRAM banks into reserved banks and shared banks. The reserved banks are exclusive to each core to provide predictable timing while the shared banks are shared by all cores to efficiently utilize the resources. MEDUSA has two separate queues for read and write requests, and it prioritizes reads over writes. In processing read requests, MEDUSA employs a two-level scheduling algorithm that prioritizes the memory requests to the reserved banks in a Round Robin fashion to provide strong timing predictability. In processing write requests, MEDUSA largely relies on the FR-FCFS for high throughput. We implemented MEDUSA in a cycle-accurate full-system simulator. The results show that MEDUSA achieves up to 91% better worst-case performance for real-time tasks while achieving up to 29% throughput improvement for non-real-time tasks 

Second, we studied the contention at shared caches and its impact on predictability. We demonstrate that the prevailing cache partition techniques does not necessarily ensure predictable cache performance in modern COTS multi-core platforms that use non-blocking caches to exploit memory-level-parallelism (MLP). Through carefully designed experiments using three real COTS multi-core platforms (four distinct CPU architectures) and a cycle-accurate full system simulator, we show that special hardware registers in non-blocking caches, known as Miss Status Holding Registers (MSHRs), which track the status of outstanding cache-misses, can be a significant source of contention. We propose a hardware and system software (OS) collaborative approach to efficiently eliminate MSHR contention for multi-core real-time systems.We implement the hardware extension in a cycle-accurate full-system simulator and the scheduler modification in Linux 3.14 kernel. In a case study, we achieve up to 19% WCET reduction (average: 13%) for a set of EEMBC benchmarks compared to a baseline cache partitioning setup. 

This dissertation investigates the feasibility of creating a multichannel sense-and-avoid radar system for small fixed-wing unmanned aerial vehicles (UAVs, also known as sUAS or drones). These aircraft are projected to have a significant impact on the U.S. economy in both the commercial and government sectors, however, their lack of situation awareness has caused the FAA to strictly limit their use. Through this dissertation, a miniature, multichannel, FMCW radar system was created with a small enough size, weight, and power (SWaP) that would allow it to be mounted onboard a sUAS providing inflight target detection. The primary hazard to avoid are general aviation (GA) aircraft such as a Cessna 172 which was estimated to have a radar cross section (RCS) of approximately 1 sqr meter. The radar system is capable of locating potential hazards in range, Doppler, and 3-dimensional space using a patent pending 2-D FFT process and interferometry. The initial prototype system has a detection range of approximately 800 m, with 360-degree azimuth coverage, and +/- 15-degree elevation coverage and draws less than 20 W. From the radar data, target detection, tracking, and the extrapolation of the target behavior in 6-degree of freedom was demonstrated.

The Treatise on Invertebrate Paleontology is the most authoritative compilation of the invertebrate fossil records. The quality of studies in paleontology, in particular depends on the accessibility of fossil data. Unfortunately, the PDF version of Treatise currently available is just a scanned copy of the paper publications and the content is in no way organized to facilitate search and knowledge discovery. This project builds an Information Retrieval based system, to extract the fossil descriptions, images and other available information from Treatise. This project is divided into two parts. The first part deals with the extraction of the text and images from the Treatise, organize the information in a structured format and store in a relational database, build a search engine to browse fossil data. Extracting text requires identifying common textual patterns and a text parsing algorithm is developed to identify the patterns and organize the information in a structural format. Images are extracted using the image processing techniques like image segmentation, morphological operations etc., and then associated with the corresponding textual descriptions. A Search engine is built to efficiently browse the extracted information and also the web interface provides options to perform many useful tasks with ease. The second part of this research focuses on the implementation of Content Based Information Retrieval System. All images from treatise are grayscale fossil images and identifying the matching images based on the visual image features is a very difficult task. Hence, we employed an approach that integrates textual and visual features to identify matching images. Textual features are extracted from the description of the fossils and using statistical approaches and Parts of Speech tagging approaches, an ontology is generated, that forms attribute – property pairs explaining how a region looks like in each shell. Popular image features like SIFT, GIST, and HOG features are extracted from fossil images. Both the textual and image features are then integrated to extract the information related to the fossil image matching the query image.

In data mining, duplicate data plays a huge role in deciding the set of rules. In this project, an analysis has been made on finding the impact of duplicates in the input data set on the rule set. The effect of duplicates is being analyzed using the error rate factor. Error rate is calculated by comparing the obtained rule set against the testing part of input data. The results of experiments have shown decrement of error rate with the increase of percentage of duplicates in the input data set, which demonstrates that the duplicate data plays a crucial role in validation process of machine learning. LEM2 algorithm and rule checker application have been implemented as a part of project. LEM2 algorithm is used to induce the rule set for the given input data set and rule checker application is used to calculate the error rate.