Defense Notices

With the motivation to develop a resilient and survivable networking system that can cope with challenges posed by the rapid growth in networking technologies and use paradigms and the impairments of TCP and UDP, we propose a general-purpose, configurable and adaptable multipath-capable transport-layer protocol called ResTP. By supporting cross- layering, ResTP allows service tuning by the upper application layer while promptly reacting to the underlying network dynamics by using the feedback from the lower layer. Our composable ResTP not only has the flexibility to provide services to different application classes operating across various network environments, its selection of mechanisms also increases the resilience level of the system in which it is deployed since the design of ResTP is guided by a set of principles derived from the ResiliNets framework. Moreover, the implementation of ResTP employs modular programming to minimize the complexity while increasing its extensibility. Hence, the addition of any new algorithms to ResTP would require only some small changes to the existing code. Last but not least, many ResTP components, including its header, are optimized to reduce unnecessary overhead. In this proposal, we introduce ResTP’s key functionalities, present some preliminary simulation results of ResTP in comparison with TCP and UDP in ns-3, and discuss our plan towards the completion and analysis of the protocol. The results show that ResTP is a promising transport-layer protocol for Future Internet (FI) resilience. 

Remote Procedure Calls (RPCs) are an integral part of the internet of things. After the introduction of RPCs, there have been a number of optimizations to amortize the network overhead, including the addition of asynchronous calls and batching requests together. In Haskell, we have discovered a principled way to compose procedure calls together using the Remote Monad mechanism. A remote monad has primitive operations that evaluate outside the local runtime system and is a generalization of RPCs. Remote Monads use natural transformations to make modular and composable network stacks which can automatically bundle requests into packets in a principled way, making them easy to adapt for a number of applications. We have created a framework which has been successfully used to implement JSON-RPC, a graphical browser-based library, an efficient bytestring implementation, and database queries. The result of this investigation is that the cost of implementing bundling for remote monads can be amortized almost for free, if given a user-supplied packet transportation mechanism.

Covert systems are designed to operate at a low probability of detection in order to provide system protection at the physical layer level. The classical approach to covert communications aims at hiding the covert signal in noise by lowering the power spectral density of the signal to a level that makes it indistinguishable from that of the noise. However, the increasing demand for modern covert systems that can provide better protection against intercept receivers (IRs) and provides higher data rates has shifted the focus to the design of Ad-Hoc covert networks (ACNs) that can hide their transmission in the RF spectrum of primary networks (PNs). The early work on exploiting the RF band of other wireless systems has been promising; however, the difficulties in modeling such environments, and analyzing the impact on/from the primary network have limited the work on this crucial subject. In this work, we provide the first comprehensive analyses of a covert network that exploits the RF band of an OFDM-based primary network to achieve covertness. A spectrum access algorithm is presented which would allow the ACN to transmit in the RF spectrum of the PN with minimum interference. Next, we use stochastic geometry to model both the OFDM-based PN as well as the ACN. Using stochastic geometry would also allow us to provide a comprehensive analysis for two metrics, namely an aggregate metric and a ratio metric. These two metrics quantify the covertness and performance of the covert network from the perspective of the IR and the ACN, respectively. The two metrics are used to determine the detectability limits of an ACN by an IR. The two metrics along with the proposed spectrum access algorithm will be used to provide a comprehensive discussion the design the ACN for a target covertness level, and analyze the effect of the PN parameters on the ACN expected performance. This work also addresses the question of trade-off between the ACN covertness and its achievable throughput. The overall research work illustrates the strong potential for using man-made transmissions as a mask for covert communications. 

The goal of this project is to bring together the complexity of core mathematics with programming abilities to code machine learning algorithms that can be incorporated into programming labs and exercises for graduate and undergraduate machine learning students. 
The aim of building the labs is to provide students with a learning tool to gain a better understanding of the inner workings of machine learning algorithms. Additionally, the labs aim to expose what challenges each algorithm can bring on its own. SAS Analytics brings into perspective machine learning methods by explaining that machine learning enables “high-value predictions that can guide better decisions and smart actions in real time without human intervention.”[2] Machine learning methods can be applied to a wide spectrum of domains and therefore, rather than attempting to cover all the algorithms, I have incorporated the algorithms that are widely applicable and explore key mathematical concepts. These algorithms for machine learning labs will give the students a learning approach to solving the intricacies of the underlying mathematical principles and will also help students to make better decisions about algorithm design and develop more accurate model predictions. 
Since each machine learning lab focuses on a particular algorithm, each program comes with a different challenge. To write these labs, I first had to master the material, which entailed finding the purpose of the algorithm and the statistical knowledge involved. Through these findings, I developed labs with specific designs, datasets, and evaluation metrics. A key difference between this approach and many other machine learning textbook approaches is that the students are building up these individual labs from scratch. They are asked to write, for a variety of different algorithms, the cost/loss function, the optimization procedure and even basic evaluation metrics. While it may be easier to call a function within a programming language, it is also easy to violate assumptions or requirements of these algorithms. By programming algorithms from scratch, as students must do in this lab, they are better able to draw parallels between the applied and theoretical underpinnings of these algorithms. 

Increase in sea level is a problem of global importance because of its impact on infrastructure and residents in coastal regions. Airborne and satellite observations have shown that the margins of Greenland and Antarctic ice sheets are melting and retreating, steadily increasing their contribution to sea level rise over the last decade. To understand the ice dynamics and develop models to generate accurate estimates of ice sheets’ future contribution to sea level rise, more information on ice thickness and basal conditions are required. Airborne ice penetrating radars are routinely deployed on long-range aircraft to perform ice thickness measurements, which are needed to derive information on bed topography and basal conditions. Acquiring useful radar reflections from the ice-bed interface is very challenging in regions where ice sheets are exhibiting the most rapid changes because returns from the ice-bed are very weak and often masked by the off nadir surface clutter. Advanced signal processing techniques, such as Synthetic Aperture Radar (SAR) and array processing, are required to filter the clutter and extract weak bed echoes buried in the noise. However, past attempts to detect these signals have not been completely successful because system and target-induced errors on SAR and array processing are not fully compensated. SAR processing in areas with significant surface slope degrades signal-to-noise ratio. Also, systematic and random errors in amplitude and phase between receive channels degrade the performance of array processors used to synthesize cross-track beam pattern. 
A novel Multi-look Time Domain Back Projection (MLTDBP) parallel processor has been developed to accurately model the electromagnetic wave propagation through the ice and generate echograms with better SNCR (Signal to Noise-Clutter Ratio) in the along-track dimension. A novel dynamic channel equalization method (based on null optimization) has been developed to adaptively calibrate the receive channels, giving an improved SNCR for the cross-track processing algorithms. Results from two-dimensional processing algorithms have been shown to be effective in extracting weak bed echoes, sloped internal ice layers, deep internal ice layers; and these results are also used to generate 3D ice-bed map of fast flowing Kangiata Nunaata Sermia (KNS) glacier in southwest Greenland.

High speed photonic systems and networks require electro-optic modulators to encode electronic signal onto optical carrier. The central focus of this research is twofold. First, tunable properties and tuning mechanisms of optical materials like Graphene, Vanadium dioxide (VO2), and Indium Tin Oxide (ITO) are characterized systematically in the 1550nm telecommunication wavelength. Then, these materials are implemented to design novel nano-photonic devices with high efficiency and miniature footprint suitable for photonic integration. 
Specifically, we experimentally investigated the complex index of graphene in near infrared (NIR) wavelength through the reflectivity measurement on a SiO2/Si substrate. The measured change of reflectivity as the function of applied gate voltage is highly correlated with the Kubo formula. Based on a fiber-optic pump-probe setup we demonstrated that short optical pulses can be translated from pump wavelength to probe wavelength through dielectric-to-metal phase transition of VO2. In this process, pump leading edge induced optical phase modulation on the probe is converted into an intensity modulation through an optical frequency discriminator. We also theoretically modeled the permittivity of ITO with different levels of doping concentration in NIR region. 
We proposed an ultra-compact electro-optic modulator based on switching plasmonic resonance “ON” and “OFF” of ITO-on-graphene via tuning of graphene chemical potential through electrical gating. The plasmonic resonance of ITO-on-graphene significantly enhances the field interaction with graphene which allows the size reduction compare to graphene based modulators without ITO. We presented a scheme of mode-multiplexed NIR modulator by tuning ITO permittivity as the function of carrier density through applied voltage. The wisely patterned ITO on top of an SOI ridge waveguide portrayed the independent modulation of two orthogonal modes simultaneously, which enhances functionality per-area. We proposed a theoretical model of tunable anisotropic metamaterial composed of periodic layers of graphene and Hafnium Oxide where transversal permittivity can be tuned via changing the chemical potential of graphene. A novel metamaterial assisted tunable photonic coupler is designed by inserting the proposed artificial tunable metamaterial in the coupling region of a waveguide coupler. The coupling efficiency can be tuned by changing the permittivity of metamaterial through electrical gating. 

Recently machine learning has been creating great strides in many areas of work field such as health, finance, education, sports etc., which has encouraged demand for machine learning systems. By definition machine learning automates the task of learning in terms of rule induction, classification, regression etc. This is then used to draw knowledgeable insights and to forecast an event before it actually takes place. Despite this automation, machine learning still does not automate the task of selecting the best algorithm(s) for a specific dataset. With the rapidly growing machine learning algorithms it has become difficult for novices as well as researchers to choose the best algorithm. The crux of a machine learning system is (1) to solve fundamental problems of preprocessing the data to help machine learning algorithm understand the data better; (2) to solve the problem of choosing meaningful features hence reducing the noise from the data; and (3) to choose the best resulting machine learning algorithm which is performed by doing grid search over hyperparameters of various machine learning algorithms and afterwards doing metric comparison amongst all outcomes. These are the problems addressed by Wolf. 
Automation is the fuel that drives Wolf. Automating time-consuming and repeatable tasks are the defining characteristics of the project. The rising scope of Artificial Intelligence (AI) and machine learning increases the need for automation to simplify the process, hence help researchers and data scientists dig deeper into the problem and understand it well, rather than spending time in tweaking the algorithms. The positive correlation of growing intelligence and the complexity of solutions has shifted the trend from Artificial Intelligence (AI) to Automated Intelligence, a paradigm on which Wolf is based. 
Wolf has been built to have an impact on a wider audience. The automation of machine learning pipeline saves ~40% of the work effort spent towards implementing and testing algorithm. It helps people with different levels of expertise and requirements, helps novices to identify best combinations of algorithms without having in depth knowledge of algorithms and helps researchers and businesses better their machine learning knowledge to figure out best resulting hyperparameters. 

As it is becoming a significant part of our daily life, wireless mobile handsets have become faster and smarter. One of the main remaining requirement by users is to have a longer lasting wireless cellular devices. Many techniques have been used to increase the capacity of the battery (Ampere per Hour), but that increases the safety concern. 
Instead, it is better to have mobile handsets that consume less energy i.e increase energy efficiency. Therefore, in this research proposal, we study and analyze the radio 
frequency(RF) transceiver energy consumption, which is the largest energy consumed in the cellular device. We consider a model of large number of parameters in order to make it more realistic. First a transmitter energy of single antenna device is considered for a fixed target probability of error in the receiver for multilevel quadratic amplitude modulations (MQAM). It will be found that the power amplifier (PA) consumes the highest portion of transceiver energy due to the low efficiency of the PA.
Furthermore, when MQAM and raised cosine filter are used, the impact of peak to average ratio (PAR) on PA becomes another source of energy wasting in the PA. This issue is analyzed in this research proposal with a number of promising solutions. This analysis of energy consumption for single antenna devices will help us analyze the energy consumption of multiple antennas devices. In this regard, we discuss the energy efficiency of multiple input multiple output (MIMO) antenna with known channel state information (CSI) at the transmitter. However, the study of energy efficiency of MIMO without CSI using space time coding will be our next step. 

IRIM (Interesting Rule Induction Module) is a rule induction system designed to induce particularly strong, simple rule sets. Additionally, IRIM does not require prior discretization of numerical attribute values. IRIM does not necessarily produce consistent rules that fully describe the target concepts, however, the rules induced by IRIM often lead to novel revelations of hidden relationships in a dataset. In this paper, we attempt to extend the IRIM system to be able to handle missing attribute values (in particular, lost and do-not-care attribute values) more thoroughly than ignoring the cases that they belong to. Further, we include an implementation of IRIM in the modern programming language Python that has been written for easy inclusion in within a Python data mining package or library. The provided implementation makes use of the Pandas module which is built on top of a C back end for quick performance relative to the performance normally found with Python. 

Data classification is a methodology of data mining used to retrieve meaningful information from data. A model is built from the input training set which is later used to classify new observations. One of the most widely used models is a decision tree which uses a tree like structure to list all possible outcomes. Decision trees are preferred for their simple structure, requiring little effort for data preparation and easy interpretation. This project implements ID3, an algorithm for building the decision tree using information gain. The decision tree is converted to a set of rules and the error rate is calculated using the test dataset. The dataset is usually stored in a relational database in the form tables. In practice, it might be desired that data be stored across multiple databases. In such scenarios, retrieving and coordinating data from the databases could be a challenging task. This project provides the implementation of ID3 algorithm with the convenience of reading data stored at multiple data sources.