The feature selection step - the ability for a model to extract good features - is the most important aspect of any model, as if some features that are part of the set which defines the concept class are excluded, no machine learning algorithm can capture and truly learn the concept class of interest. It has been the norm that the feature selection step is carried out on groups of properties, called attributes. In this work, we propose a new method for feature selection based on Information Gain, that allows for selection of attributes without assuming independence between them and by assigning ranks to properties instead of the whole attribute. This approach can be used with single class learning given that the probability distribution of the features is known. We tested this approach against Mutual Information and TFIDF, using a model based on conceptual space theory. Despite the mostly biased model, a slight increase in the accuracy of classification has been obtained compared to the other feature selection methods. Based on the dataset used, we found out that the absence of certain instructions such as jmp, could be used to detect Trojans with weak association. We anticipate that the proposed approach can be used in future work to allow a model to construct its own dimensions freely from sequences of opcodes.
A novel Doppler LiDAR sensor has been developed with the help of NASA technology transfer. This sensor can make highly accurate, three-dimensional velocity measurements in the body frame of a vehicle. However, the practicability of such a sensor in an IMU-mechanized navigation solution is unknown. A covariance analysis can be run with a simulation flight across multiple trajectories as well as different IMU grades to understand the sensors accuracy and contribution to a navigation system. With a full simulated analysis of this sensor, it is much easier for researchers and others to understand the full benefits that a Doppler Lidar system might bring to a navigation solution. We conduct the analysis by way of covariance analysis as well as Monte Carlo and give a brief explanation of what covariance analysis is and how it can be used to obtain results for this new sensor. We define this new software tool we have created and speak on how researchers can use it to conduct covariance analysis tests of their own.
A microsphere lens integrated on top of an MWIR photodetector has exhibited an increase in detector overall sensitivity over a desired spectral bandwidth and, more importantly, a decrease in Noise-to-Signal-Ratio (NSR), as we have shown in our prior work. Using signal processing methods, we have also shown, for the first time, the existence of an adverse effect of microsphere misalignment on the detector's overall sensitivity. In this paper, we model and analyze the effect of various misalignment types between the microsphere-lens and the photodetector on the resulting detector sensitivity. As we show here, microsphere lens-sensor misalignment can result, to some varying degrees, in reduced spectral responses in microspheres-lens-enhanced MWIR photodetectors. We also model and analyze the effect of such misalignment on multiple SLS photodetector sizes, namely 35 $\mu\mathrm{m}$ and 40 $\mu \mathrm{m}$, and microsphere material types, namely Sapphire and Polystyrene. In the case of Polystyrene material, the results show that the sensitivity of the microsphere-lens-enhanced detector is mainly and significantly reduced by the microsphere Polystyrene material alone around wavenumber 3000, due to the Polystyrene material spectral absorptions around that wavenumber.
In this paper, the real-time cooperative guidance strategy for the aircraft trying to evade an incoming attacking missile was developed. The aircraft performs evasive maneuvers, launching a defender missile to divert the missile. Instead of classical strategies, which are based on optimal control or differential game, the engagement was formulated as a multi-agent game. The problem was solved by using the deep reinforcement learning method. To address the sparse reward problem, a general reward design method was presented utilizing the shaping technique. Guidance law, reward function, and training approach were demonstrated through the learning process and simulations. The application of the non-sparse reward function accelerated the model convergence. Considering a standard optimal guidance law as a benchmark, the effectiveness and advantages (guarantees of the aircraft's escape and win rates in multi-agent game) of the proposed guidance strategy were validated by the simulation results. Compared to the standard optimal guidance law, the proposed guidance strategy performed better with less prior knowledge.
This paper presents an octree-based compression algorithm for LiDAR point cloud data. This solution is presented as an alternative to LASzip, the current state-of-the-art in LiDAR compression, that provides an average 60% higher compression ratio on Geiger-mode data, and is capable of natively compressing both LAS and BPF file formats.
Collision avoidance is a crucial task in vision-guided autonomous navigation. Traditional solutions tend to be computationally expensive and difficult to adapt to new environments. In this work, we propose a novel collision avoidance solution for autonomous drones. Formulated under a deep reinforcement learning framework, our model relies on a pair of margin reward functions to ensure the drones fly smoothly while greatly reducing the chance of collision. Additional reward functions are designed to attract the drones to fly towards their destinations, as well as to follow predefined routes. Experiments using indoor simulation environments demonstrate the effectiveness of our overall design and the individual components.
Waveform-based speech enhancement solutions have become increasingly popular in recent years. They commonly take a certain fully convolutional network (FCN) architecture that relies on convolutions to capture discriminative features at multiple scales. We found FCNs with standard convolutions tended to overfit on training data, likely due to the large number of trainable parameters brought by channel integration operations. Handling short speech frames also poses a challenge for FCNs, as the context to be observed is often limited, resulting in boundary discontinuities in the concatenated outputs. In this work, we propose remedies to address the aforementioned practical issues. With the Wave-U-Net as the baseline model, we replace the standard convolutions with depthwise and depthwise separable convolutions to compress the FCN models. With the reduced model complexity, such replacements lead to significantly improved network efficiency and generalization. To address the short frame issue, we propose to utilize RNN to connect depthwise FCNs, allowing temporal information to be propagated along the networks on individual frames. Our FCN + RNN model demonstrates an excellent smoothing effect on short frames, enabling speech enhancement systems with very short delays. The effectiveness of the proposed models is validated with experiments on AzBio sentences and VCTK datasets.
This paper presents a new general proof that the roots of the polynomial corresponding to the minimum variance filter computed using the true Toeplitz covariance matrix must fall on the unit circle (UC). Unlike a previous proof applicable to the Minimum Variance Distortionless Response (MVDR) case only, the new proof does not rely on Wiener-Khinchin theorem to map the problem into frequency domain. Furthermore, the proof is applicable to a general class of sensor array signal processing problems beyond MVDR. Next, new closed-form solutions of UC roots constrained (UCRC) MVDR beamformer and Adaptive Matched Filter (AMF) will demonstrate significant performance improvement than the conventional, non-UC counterparts. The proposed approach will also be shown to achieve superior performance improvement at smaller aperture sizes than the conventional approaches, justifying its suitability for applications with reduced SWAP requirements.
In statistics, regression analysis includes many techniques for modeling and analyzing several variables, when the focus is on the relationship between a dependent variable and one or more independent variables. More specifically, regression analysis helps one understand how the typical value of the dependent variable changes when any one of the independent variables is varied, while the other independent variables are held fixed. Most commonly, regression analysis estimates the conditional expectation of the dependent variable given the independent variables that is, the average value of the depen-dent variable when the independent variables are fixed. Less commonly, the focus is on a quaintly, or other location parameter of the conditional distribution of the dependent variable given the independent variables. In all cases, the estimation target is a function of the independent variables called the regression function. In regression analysis, it is also of interest to characterize the variation of the dependent variable around the regression function, which can be described by a probability distribution.
Human subjects' identification including face recognition, fingerprint recognition, and gait recognition enhances biometric health and safety. However, existing methods have their limitations as it is difficult to identify humans when humans cannot touch devices or there is a dim environment. This paper proposes a novel human identification approach, which utilizes passive radio frequency (RF) signal as a biometrics modality to achieve human identification. The passive human subject identification with radio-frequency (PHSIR) approach is verifies that different human subjects could generate different spectrum signatures, and these spectrum characteristics can be distinguished by machine learning (ML) algorithms to achieve human subjects' classification. Software-defined radio (SDR) technology acquires passive RF in the frequency bands that are sensitive to human occupancy. The passive spectrums were collected in two environments. Four ML algorithms were used to classify the sample spectrums associated with different human subjects, including decision tree, support vector machines (SVM), k-nearest neighbors (KNN), and random forest. Experimental results from seven volunteers indicate the classification accuracy is higher than 94% for seven volunteers using KNN algorithms.
The utilization of optimization algorithms to allow interference direction finding with nondirectional sensor power reception in global positioning system (GPS) navigation applications is explored, and experimentation with simulated interference setups is conducted. Dimension reduction techniques are employed and evaluated to increase the efficiency of the algorithms.
The growth of wireless communications devices increased the needs for higher performance and higher density components. RFIC and packaging provides low cost solution to the future products. The quality of the interconnects connections determines the performance. This paper introduces a technique developed at National Instrument to boost the quality of the connections. The technique is based on Hot Vias interconnects for 60 GHz MMICs using silver epoxy with minimum added cost for the manufacturing process.
Small Flapping-Wing Micro-Air Vehicles (FW-MAVs) can experience wing damage and wear while in service. Even small amounts of such damage require significant adaptation of on-board flight control to maintain flight path precision. Previous work employed a custom Evolutionary Algorithm (EA) that adapted wing motion patterns, while in flight and in normal online service, to compensate for wing damage. Although successful in finding solutions to this challenging online non-stationary problem, the previous methods would often require hours of flight time to reach full success. This paper details a completely new EA and wing motion basis function set that cuts the learning time to on average four minutes while not sacrificing final controller quality. The paper will provide both conceptual discussion of how the new algorithm delivers such dramatic improvements and extensive simulation results demonstrating robustness and consistently low time requirements over a wide variety of wing damage conditions.
Binary translation (BT) is the process of converting executable binary from one instruction set architecture (ISA) to another. Accelerated binary translation (XBT) refers to BT using FPGA for hardware acceleration and feeding the target processor at-speed. This work proposes a reconfigurable pipelined structure built on FPGA that performs XBT on different ISAs. An XBT system that translates MIPS to RISC-V is implemented and tested on the Xilinx Zynq platform. Results of several benchmarks show obvious speedup of approximately 48 times compared to an equivalent software approach.
In this paper, the propagation of plane electromagnetic (EM) waves across a boundary between a lossless achiral dielectric and a lossy chiral medium within specific band of chirality coefficients has been investigated. Standard Fresnel coefficient (FC) calculations are followed by derivation of the reflectance and transmittance across the interface. The results are compared to previous studies on EM propagation across an achiral/chiral (ACC) interface based on amplitude analysis.
Recent literature work has shown that many machine learning models are vulnerable to adversarial attacks, but there also exist techniques that can be used to improve their robustness. Here, we apply some adversarial attack methods for the purpose of generating adversarial features. The said adversarial attacks are directed towards a neural network that is initially trained to classify malware programs in the labeled subset of the BIG 2015 dataset based on the malware's re-sampled and resized HEX codes, which thereby constitute the aforementioned features. Furthermore, via adversarial training, we investigate the creation of a more robust neural network.
We present a brief overview of a temporal hypergame framework that can be used to model classic game theoretic principles, in addition to games where there is a difference in perception. This paper discusses the application of the temporal hypergame framework to the classical game theoretic iterated Prisoner's Dilemma. Using the temporal hypergame framework, the concept of backwards induction is shown as a solution to the Prisoner's Dilemma. The Prisoner's Dilemma is symmetric and does not have differences in perception, therefore the structural properties do not allow for detailed hypergame analysis. Next, the temporal hypergame framework is applied to an attacker-defender network game. The attacker-defender network game is used to show how the framework can be applied under conditions of difference in perception. This examples used to show validity to the classical game theoretic problems to the hypergame framework.
In this study, THz GaN bow-tie antennas for space and defense applications are designed and their performance are analyzed using the multiphysics program, Comsol. GaN is attractive for its radiation hardness, resilience to abrasive atmospheres, thermal stability, and ability to integrate electrical and optoelectronic devices. The results show that the device performance is dependent on the materials used and the device geometry. The study also shows how to optimize an antenna array for the required frequency band.
A low power, small area front collision avoidance circuit using Light Detection and Ranging technology is presented in this paper. The proposed system would help detect and avoid the objects which would otherwise collide with an autonomous vehicle. After the front sensor detects stopped or slowed down vehicles, all sensors will turn on and make decisions based on the system's algorithm calculations. The system is implemented in CMOS 90nm technology. The power consumption of the system is 1.424mW and uses a 1.2V power supply. 4ns resolution front end sensors and 1ns resolution side sensors are used to compare the target distance with a safety standard. The processing time for making decisions is 889.5ns when the car is traveling 45mph which includes the 800ns sensor detection time.
This paper presents the development of a low-SWaP spiking symbolic Bayesian network reasoner that makes decisions based on given state inputs. The Bayesian network is demonstrated onboard an unmanned aerial vehicle using the Intel Loihi, a neuromorphic research hardware, and achieves improved power efficiency, runtime performance, and functional equivalency when compared to conventional hardware.
