College of Engineering and Computing
Permanent URI for this community
Browse
Browsing College of Engineering and Computing by Title
Now showing 1 - 20 of 150
Results Per Page
Sort Options
Item A CNN/MLP Neural Processing Engine, Powered by Novel Temporal-Carry-deferring MACs(2021) Mirzaeian, AliThe applications of machine learning algorithms are innumerable and cover nearly every domain of modern technology. During this rapid growth of this area, more and more companies have expressed a desire to utilize machine learning techniques in smaller devices, such as cell phones or smart Internet of Things (IoT) instruments. However, as machine learning has so far required a power source with more capacity and higher efficiency than a conventional battery. Therefore, introducing neural network accelerators with low energy demands and low latency for executing machine learning techniques has drawn lots of attention in both the academia and industry.Item A Literature Review of Network Models in Neuroscience(2022-05) Kazemi Abharian, Sanaz; Parsa, Maryam; Peixoto, NathaliaNeuroscience is developing in size, scope, and complexity of neural data obtained from large portions of nervous systems. The primary aim of network models in neuroscience is to map, record, and model the components and interactions of the human brain and nervous systems. In this paper, our goal is to survey graph-based network models in neuroscience and study their applications to detect diseases and disorders related to the human brain.Item A Literature Review of the Development Trends of Visual Neural Protheses(2021) Givens, JordanAccording to a systemic review of population-based data sets relevant to global vision impairment and or blindness between 1980 in 2015, there is an estimated 36 million people who suffer from blindness [4]. An additional 405.1 million people with mild to severe visual impairment [4]. Visual impairment of any degree and its growing prevalence are not a new issue. However, recent advancements in neural protheses, such as cochlear implants that aid those who are hearing impaired, leading researchers to turn to visual neural protheses. Visual neural prostheses focus on the concept of artificially inducing vision by using our current understanding of electrical stimulation, visual pathways, and visual sensations. Therefore, all visual protheses focus on creating an artificial sense of vision through the electrical activation of neurons belonging to the visual system of the body [2]. There are a variety of approaches researchers have taken to accomplish this, the approaches differ in the aspects of the visual system they attempt to replace. As of 2019 they are primarily four approaches that encompass most neural prosthetics. Those that focus on the retina, optical nerve, cortical region of the brain, and or the lateral geniculate nucleus (LGN) within the thalamus [2][3]. The most prominent of which is the retina neural prosthetic, having multiple current implants such as the Argus II electronic epiretinal device, this is due to its extracranial location and simpler organization compared to other methods [3][5]. However, the retinal approach is not without faults of its own such as unwanted electrochemical reactions and low resolution. Therefore, in this paper the four approaches of visual neural prostheses will be examined to provide a greater insight into the field.Item A machine learning approach to predict rtms therapy response in major depressive disorder(2020-05) Shams, MohammadMachine learning techniques have been utilized to predict the outcome of repetitive transcranial magnetic stimulation (rTMS) treatment in depression, e.g., through classifying the responders (R) and non-responders (NR) to rTMS treatment for major depression disorder (MDD) patients. MDD is among the leading causes of disability in the world with affecting more than 260 million people, and a major contributor to the overall global burden of disease. In this study, the outputs of the Local Subset Feature Selection (LSFS) method were used by an SVM classifier to evaluate the capability of the proposed method in the prediction of rTMS treatment response in depression cases. A Leave-One-Out cross-validation method is applied to the input data to evaluate the performance of the response classification. The achieved accuracy, sensitivity, and specificity were 89.5%, 90%, and 87%, respectively. The main restriction of this study that would limit its usage in clinical applications is the small sample size.Item A Novel Application of Musculoskeletal Ultrasound Imaging(Journal of Visualized Experiments, 2013-09) Eranki, Avinash; Cortes, Nelson; Ferenček3, Zrinka Gregurić; Siddhartha, SikdarUltrasound is an attractive modality for imaging muscle and tendon motion during dynamic tasks and can provide a complementary methodological approach for biomechanical studies in a clinical or laboratory setting. Towards this goal, methods for quantification of muscle kinematics from ultrasound imagery are being developed based on image processing. The temporal resolution of these methods is typically not sufficient for highly dynamic tasks, such as drop-landing. We propose a new approach that utilizes a Doppler method for quantifying muscle kinematics. We have developed a novel vector tissue Doppler imaging (vTDI) technique that can be used to measure musculoskeletal contraction velocity, strain and strain rate with sub-millisecond temporal resolution during dynamic activities using ultrasound. The goal of this preliminary study was to investigate the repeatability and potential applicability of the vTDI technique in measuring musculoskeletal velocities during a drop-landing task, in healthy subjects. The vTDI measurements can be performed concurrently with other biomechanical techniques, such as 3D motion capture for joint kinematics and kinetics, electromyography for timing of muscle activation and force plates for ground reaction force. Integration of these complementary techniques could lead to a better understanding of dynamic muscle function and dysfunction underlying the pathogenesis and pathophysiology of musculoskeletal disorders.Item A Proposal for a W3C XG on Uncertainty Reasoning for the World Wide Web(Information Extraction & Transport, Inc. (IET), 2006-11) Laskey, Kenneth J.; Costa, Paulo C. G.; Laskey, Kathryn B.The Semantic Web envisions effortless cooperation between humans and computers, seamless interoperability and information exchange among web applications, and rapid and accurate identification and invocation of appropriate Web services. At the current stage of evolution in Semantic Web research, there is a growing understanding that a major step towards this vision involves the implementation of principled uncertainty representation and reasoning in SW applications. This position paper introduces initial thoughts on how the World Wide Web Consortium (W3C) Incubator XG process could be employed to move forward the concept of a Web with uncertainty.Item A Review of the Effects of Microwave Radiation on Spatial Memory and Learning(2022-05) Dockum, Allison; Peixoto, NathaliaMicrowave radiation refers to electromagnetic waves between 300 MHz and 300 GHz. Microwaves are used in communication systems, manufacturing, medical treatments, and military operations. The ability to modify waveform parameters, such as frequency and duty cycle, contribute towards the versatility of microwave radiation. In the same way, the ability to adjust each parameter also contributes to the complexity of understanding the biological effects of microwaves. For some time, researchers have studied the microwave effects on learning and spatial memory in rodents. Rodents provide scientists with a neurologically similar model to humans, which is easy to study both on the cellular levels and to assess behaviorally due to the development of maze performance tests. However, the mechanisms disrupting spatial memory remain largely unknown because of the nearly infinite number of ways microwave can be modified, combined with the multitude of neurological effects, which could impact behavior. It is important for scientists to continue to study rodent models under various microwave exposure conditions to prevent harmful exposure conditions with humans. In this review microwave exposure conditions will be introduced, followed by an introduction to the cruciality of synaptic plasticity to spatial memory and learning. Synaptic plasticity can be impacted through various neurological mechanisms; of which the NMDAR receptors, neurotransmitter release, and activation of intracellular signaling cascades, and cell apoptosis will be reviewed.Item A Side Channel Delay Analysis for Hardware Trojan Detection(2020-10) Vakil, AshkanThis research proposal introduces a learning assisted modeling technique for the purpose of Hardware Trojan detection. Our proposed model, unlike the prior art, does not require a Golden fabricated chip as a fingerprint to compare the side channel signals. Instead, by modeling the voltage drop and voltage noise pre-fabrication, and with training a Neural Network post-fabrication, our proposed technique can improve the timing model collected during timing closure and produces a Neural assisted Golden Timing Model (NGTM) for side channel delay-signal analysis. The Neural Network acts as a process tracking watchdog for correlating the static timing data (produced at design time) to the delay information obtained from clock frequency sweeping test. Proposed modeling technique enables Hardware Trojan detection close to 90% in the simulated scenarios.Item A Study of Epileptic Seizure Detection using Machine Learning Algorithms(2022-05) Kamaraju, Rajeev; Peixoto, NathaliaThis paper focuses on studying epileptic seizure detection using machine learning algorithms. Algorithms like Naïve Bayes, Logistic Regression, Stochastic Gradient Descent, KNearest Neighbour, Decision trees and random forests have been studied. For each of the classifier, many performance metrics have been computed and Area Under Curve (AUC) has been chosen as our performance metric. The paper also introduces the possibility of detecting epileptic seizures using Neural networks.Item A Survey On Techniques Used for Designing Fault Tolerant and Process Variation Aware Memories and Caches(2020) Beheshti-Shirazi, Sayed AreshAggressive voltage and frequency scaling schemes applied to memory and cache structures, specially for memory systems fabricated in advanced and scaled geometry nodes that are severely affected by process variation,significantly increases the likelihood of read, write and access failures to/from memory cell array, and reduces the extent of frequency and voltage scaling. To remedy this problem, in the past decade, many researchers have investigated alternative and fault tolerant cache and memory organizations to mitigate the impact of process variation, and to reduce the failure rate of memory array in the results of voltage and frequency scaling. This paper discusses and compare many of such cache and memory design techniques.Item A Technical Report On Accelerator Design For Deep Neural Networks(2019) Mirzaeian, AliItem A Technical Report on Key Management and Vulnerability Assessment of Logic Obfuscation(2019) Zamiri Azar, KimiaItem A Technical Report on Logic Obfuscation using Reconfigurable Logic and Routing Blocks(2019) Mardani Kamali, HadiThe increasing cost of building, operating, managing, and maintaining state-of-the-art silicon manufacturing facilities has pushed several stages of the semiconductor device’s manufacturing supply chain offshore. However, many of these offshore facilities are identified as untrusted entities. Processing and fabrication of ICs in an untrusted supply chain poses a number of challenging security threats such as IC overproduction, Trojan insertion, Reverse Engineering, Intellectual Property (IP) theft, and counterfeiting. To counter these threats, various hardware design-for-trust techniques have been pro- posed. Logic locking, as a proactive technique among these techniques, has been introduced as a technique that obfuscates and conceals the functionality of IC/IP using additional key inputs that are driven by an on-chip tamper-proof memory. Shortly after introducing the primitive logic locking solutions, a very strong Boolean attack, the Satisfiability (SAT) attack. It was shown that the SAT attack could break all previously proposed primitive locking mechanisms in almost polynomial time. To thwart the strength of SAT attack, researchers have investigated many directions, such as formulating locking solutions that significantly increase the number of required SAT iterations, or formulating the locking solutions such that it is not translatable to a SAT problem. However, further investigations demonstrated that some of these locking techniques are vulnerable to other types of attacks such as Signal Probability Skew (SPS) attack, removal attack, approximate-based attack(s), bypass attack, and Satisfiability Module Theories (SMT) attack. In addition, these techniques suffer from very low output corruption. Hence, an unactivated IC behaves almost identical to an unlocked IC with exception of one or few inputs.Item A Technical Report on Real-Estate Rent Prediction(2017) Rafatirad, SetarehReal-estate rent prediction is sensitive to several independent parameters and has allured a lot of researchers in the past few years to constructing automated tools using (ML) commodities. However, most of the proposed solutions are limited in scope, and are only investigated on a particular locality, house type, or based on one type of machine learning algorithm. Furthermore, the past work often used synthetic data which can compromise the accuracy of the output, as it is not closely identical to real-world datasets. To address these challenges, we study a wide range of Machine Learning techniques applied to three real-estate housing types, using real-world data. Unlike prior work which attempt to develop a one-size-fits-all model with fixed set of features, our study shows that the important parameters for rent prediction depends highly on the type and locality. Further, for each property type, there is a different winning algorithm to perform rent prediction. Accordingly, we construct multiple rent prediction models using a large Zillow dataset of 50K real estate properties in the state of Virginia and Maryland. In addition to Zillow, external attributes such as walk/transit score, and crime rate are collected from online sources. Our comprehensive case study indicates that real-estate rent behavior strongly depends on the type of house and locality. As such, we deploy a two-layer clustering approach to partition data into multiple training sets based on house-type and similar zip codes. We evaluate and report the performance of the prediction models studied in this work based on two metrics of R-squared and Mean Absolute Error, applied on unseen data.Item A Technical Report on Security Evaluation of Logic Locking: Attacks and Defenses(2020) Zamiri Azar, KimiaItem An Application of Bayesian Networks to Antiterrorism Risk Management for Military Planners(2005-11-18T20:51:11Z) Hudson, Linwood D.; Ware, Bryan S.; Laskey, Kathryn B.; Mahoney, Suzanne M.Recent events underscore the need for effective tools for managing the risks posed by terrorists. Assessing the threat of terrorist attack requires combining information from multiple disparate sources, most of which involve intrinsic and irreducible uncertainties. This paper describes Site Profiler® Installation Security Planner, a tool initially built to assist antiterrorism planners at military installations to draw inferences about the risk of terrorist attack. Site Profiler applies knowledge-based Bayesian network construction to allow users to manage a portfolio of hundreds of threat/asset pairs. The constructed networks combine evidence from analytic models, simulations, historical data, and user judgments. Site Profiler was constructed using our generic application development environment that combines a dynamically generated object model, a Bayesian inference engine, a graphical editor for defining the object model, and persistent storage for a knowledge base of Bayesian network fragment objects. Site Profiler's human-computer interaction system is tailored to mathematically unsophisticated users. Future extensions to Site Profiler will use data warehousing to allow analysis and validation of the network’s ability to predict the most effective antiterrorism risk management solutions.Item An Investigation of Machine Learning Techniques for Use in Training Agents for Military Simulations(2006-05-05T14:27:34Z) Hieb, Michael R.; Pullen, J. MarkAgents assist users with performing tasks in computer-based applications. The current practice of building an agent involves a developer programming it for each task it must perform, but agents constructed in this manner are difficult to modify and cannot be trained by a user. Agent- Disciple is a system for training instructable agents through user-agent interaction. In Agent-Disciple a user trains an instructable agent through the interface of the user’s application by providing specific examples of tasks and their solutions, explanations of these solutions and supervises the agent as it performs new tasks. We report here on our work that uses Agent-Disciple to provide a learning agent that can command simulated military forces. Military simulations currently have many limitations in modeling human behavior. While it is relatively straightforward to build models of doctrine, it is difficult to have agents utilize this doctrine in varying contexts. There are too many factors to consider when building deterministic models of behavior, even in well-defined situations. We applied Agent-Disciple to circumvent this problem by using heuristic learning methods. A case study is presented in developing an instructable Company Commander Agent for the Modular Semi-Automated Forces (ModSAF) simulation – a state-of-the-art, real-time, distributed interactive military simulation currently utilized in large-scale training exercises. A ModSAF user can train the Company Commander Agent interactively, using the ModSAF interface, to perform various military missions using the Captain system based on Agent-Disciple. A training session with the agent illustrates the different types of learning interactions available in Agent-Disciple.Item Analysis of Social Media Comments(2021-04-28) Mohamed, MazenFighting the COVID-19 is widespread acceptance of the covid-19 vaccines. Achieving the widespread uptake might be challenging and may be obstructed by the misinformation prevalent in social media. The social media platforms have become a common source of information and disinformation on vaccines. Vaccine’s hesitancy is more prevalent in social media, especially Twitter. Machine learning models can explain the social media comments about this topic.Item Attempted Prediction of Emotional Valence from EEG Using Multidimensional Directed Information(2022-05) Clayton A Baker; Peixoto, NathaliaQuantitative measurement of a person’s emotional state can aid performance in a number of areas, such as human-machine interactions, and psychological research. Electroencephalogram (EEG) data has shown potential as a predictor of emotional valence based on asymmetric activation patterns between the left and right hemispheres of the prefrontal cortex. Multidimensional directed information (MDI) is a computational tool that allows the measurement of information content transferred between different signals in a connected system, and has previously seen applications in EEG-based affective measurement in order to detect the presence of an emotional response. This study aimed to use MDI with EEG data from published datasets in order to derive a directional bias metric as a predictor for emotional valence based on frontal hemisphere asymmetry. Two methods of MDI computation were attempted; significant differences were observed in results between the two, suggesting possible errors in implementation. Neither method yielded output correlating with valence.Item Bayesian ontologies in AI systems(2006-07-30T03:10:44Z) Costa, Paulo C. G.; Laskey, Kathryn B.; AlGhamdi, GhaziOntologies have become ubiquitous in current-generation information systems. An ontology is an explicit, formal representation of the entities and relationships that can exist in a domain of application. Following a well-trodden path, initial research in computational ontology has neglected uncertainty, developing almost exclusively within the framework of classical logic. As appreciation grows of the limitations of ontology formalisms that cannot represent uncertainty, the demand from user communities increases for ontology formalisms with the power to express uncertainty. Support for uncertainty is essential for interoperability, knowledge sharing, and knowledge reuse. Bayesian ontologies are used to describe knowledge about a domain with its associated uncertainty in a principled, structured, sharable, and machine-understandable way. This paper considers Multi-Entity Bayesian Networks (MEBN) as a logical basis for Bayesian ontologies, and describes PR-OWL, a MEBN-based probabilistic extension to the ontology language OWL. To illustrate the potentialities of Bayesian probabilistic ontologies in the development of AI systems, we present a case study in information security, in which ontology development played a key role.