Browsing by Author "Pan, Miao"
Now showing 1 - 20 of 57
- Results Per Page
- Sort Options
Item A Hierarchical Game Framework for Distributive Resource Allocation in Future Heterogeneous Network(2017-12) Zhang, Huaqing; Han, Zhu; Pan, Miao; Nguyen, Hien Van; Cai, Lin X.; Niyato, DusitThe explosive development of mobile data service makes our lives convenient and efficient. However, with the increasing demands of wireless data transmission, it is difficult to fulfill real-time requirements of mobile users in the traditional cellular architecture. In future wireless communication, the network is expected to be heterogeneous. On one hand, the large amount and different sizes of small cells are expected to be overlaid within the wireless network. On the other hand, wireless networks will be coordinated with other networks for expansion of available resources. Nevertheless, due to the distributive behaviors of multiple individuals in the heterogeneous network (HetNet), it is challenging to adopt resource allocations to achieve stable and high quality of service (QoS) for all mobile users. In this dissertation, we overview the development of wireless networks and summarize the wireless service into a 4-layer service architecture, consisting of the service layer, resource layer, infrastructure layer and user layer. Considering the heterogeneous architecture of future wireless network, a hierarchical game framework is proposed to determine distributive strategies for high performance and equilibrium solutions. We first analyze the distributive behaviors during the cooperation of multiple infrastructure providers, and propose a zero-determinant strategy for the administrator of the cooperation to maintain a high social welfare. Then, we analyze the distributive behaviors of multiple resource providers as well as infrastructure providers, with the applications of LTE unlicensed (LTE-U) and visible light communication (VLC). In LTE-U, multi-leader multi-follower Stackelberg game is employed among operators and users for resource management of licensed spectrum and unlicensed spectrum. In VLC, we combine VLC with Device-to-Device (D2D) communication and employ the Stackelberg game with a graphical game to analyze the equilibrium behaviors of all individuals. Finally, we consider the general heterogeneous network with the application of fog computing. With network virtualization, a hierarchical game framework combining the Stackelberg game and matching game are applied, where each mobile user is allocated with the optimal amount of computing resources from the selected fog node or cloud server.Item A NONPARAMETRIC BAYESIAN FRAMEWORK FOR MOBILE DEVICE SECURITY AND LOCATION BASED SERVICES(2013-12) Nguyen, Nam T. 1979-; Han, Zhu; Zheng, Rong; Ogmen, Haluk; Prasad, Saurabh; Pan, MiaoIn June 2013, it was reported that, for the first time, more than half of American adults have smartphones [1]. Smartphones are carried by the users most of the time and used to access all types of personal sensitive information, from email, Facebook to banking, files server, etc. The fact that humans are more and more attached to their phones poses both good and bad aspects, namely, security challenges and new opportunities in improving users experiences. From the bad aspect, it is of interest to find out how to protect the users from cyber attack. Whereas from the good aspect, we can improve users’ experiences by providing services related to their locations. To address the first problem, in this dissertation, we propose a security framework to detect two kinds of attacks, the Masquerade attack and the Sybil attack. Most existing literature employs supervised learning and assumes the number of devices is known. We, on the other hand, propose a non-parametric Bayesian method to detect the number of devices as well as determine which observations belong to which devices in an unsupervised passive manner. An attack can be detected by comparing the number of registered users with the number of devices found, and the malicious nodes are found based on the labels of their observations. For the second problem, we propose a location based service (LBS) enabler framework by providing a high accuracy indoor location identification and future location prediction algorithms. LBS are applications in which, locations of users are utilized to activate a set of services which significantly improve users experiences. Examples include a micro-climate control application, which can automatically adjust room temperature given that the room is occupied. It also can be a network scheduling users’ access application, where users’ future whereabouts can be predicted and used for arranging files transfer to better enhance users’ experiences. In this dissertation, we mainly focus our research on the above two fields. The nonparametric Bayesian framework was used as the generative model for both the observations extracted from the wireless signal in the wireless security problem, as well the observations extracted from the features that represent a location in LBS. Beside the framework, the major contributions of the dissertation include a missing data handling algorithm, a light-weight indoor place identification algorithm, a stopping rule to terminate the algorithm in a quickest way while maintaining a acceptable false alarm rate, and a passive approach to defend against Masquerade and Sybil attacks in wireless networks. Moreover, several mechanisms to predict users’ future whereabouts such as a Dynamic Hidden Markov Model that can evolve itself over time, or a prediction model based on Deep Learning were proposed. Most of the algorithms are evaluated using experimental data and proved to obtain considerably high performances compared with other state-of-the-art approaches.Item A System-Theoretic Investigation of Hormone Dynamics in Chronic Fatigue Syndrome, Fibromyalgia Syndrome, and Obesity(2020-05) Pednekar, Divesh Deepak; Faghih, Rose T.; Pan, Miao; Reddy, Rohith K.Fibromyalgia syndrome (FMS), chronic fatigue syndrome (CFS), and obesity are complicated medical disorders with little known etiologies. The purpose of this research is to characterize FMS, CFS, and obesity by studying the variations in hormonal secretion patterns, timings, amplitudes, the number of underlying pulses, as well as infusion and clearance rates of hormones such as cortisol, and leptin. Employing a physiological state-space model with plausible constraints, we estimate the hormonal secretory events and the physiological system parameters (i.e., infusion and clearance rates). The first outcome of our research shows that the clearance rate of cortisol is lower in FMS patients as compared to their matched healthy individuals based on a simplified cortisol secretion model. Moreover, the number, magnitude, and the energy levels of cortisol secretory events are lower in FMS patients. During early morning hours, the magnitude and the energy levels of the cortisol secretory events are higher in CFS patients. Due to the lower cortisol clearance rate, there is a higher accumulation of cortisol in FMS patients as compared to their matched healthy subjects. As the FMS patients accumulate higher cortisol residues, internal inhibitory feedback regulates the hormonal secretory events. Therefore, the FMS patients show a lower number, magnitude, and the energy levels of hormonal secretory events. Though CFS patients have the same number of secretory events, the secretion quantity is lower during early morning hours. When we compare the results for CFS patients against FMS patients, we observe different cortisol alteration patterns. In the second part of this thesis, we propose a simplified minimal leptin secretion model and study the correlation between estimated parameters of leptin and cortisol. The hunger hormone leptin and stress hormone cortisol are closely associated with obesity. Traditionally, a leptin-cortisol antagonism is observed in obese patients. We also observe a leptin-cortisol antagonism when we compare the reconstructed leptin and cortisol levels, hence, further validating the model. The proposed model can potentially be employed to study leptin variations in obese patients against their matched healthy subjects. Characterizing CFS, FMS, and obesity based on the hormonal alterations will help us develop effective methods for treating these disorders.Item Algorithms for Particle Swarms Using Global Control: Aggregation, Mapping, Coverage, Foraging, and Shape Control(2017-05) Viswanathan Mahadev, Arun; Becker, Aaron T.; Tsekos, Nikolaos V.; Pan, MiaoTargeted drug delivery is a promising technique to reduce the side effects of drugs by delivering them in concentrated doses using large swarms (10^16) of controllable microbots only targeting bad or infected tissue. A promising way to control small steerable microbots is by using a global control field such as the magnetic gradient of an MRI machine. In this work we develop benchmark algorithms for performing aggregation of microbots using global control. Using our findings we develop algorithms for a novel approach of mapping tissue and vascular systems without the use of harmful contrast agents in an MRI. In our work we consider a swarm of particles in a 1D, 2D, and 3D grids that can be tracked and controlled by an external agent thus building a map. We present algorithms for controlling particles using global inputs to perform: (1) Mapping, i.e., building a representation of the free and obstacle regions of the workspace; (2) Foraging, i.e., ensuring that at least one particle reaches each target location;and (3) Coverage, i.e., ensuring that every free region on the map is visited by at least one particle. Finally we also demonstrate shape control of large swarms using global control by developing an algorithm for position control.Item Applications of Magnetic Induction and Localization to an Autonomous Underwater Vehicle(2021-04-01) Khan, Sajid; Soto, StebanAutonomous robotics has become widely popular for various industrial applications. For underwater applications, there are some issues with communication due to the high signal attenuation in water. Through prior experiments, Magnetic Induction (MI) has shown to have promising result in underwater applications with low power consumption. Localization is the one idea in robotic navigation and is done through the creation of a copper coil in the XYZ directions. Using these 3 coils, a robot’s position can be precisely determined relative to the coil and a network of these coils can be created to collect data on the robotics location in the water. This research explores the application of MI to autonomous vehicles, in hopes of being able to create an underwater network of Autonomous Underwater Vehicles (AUV) which would be able collect data and deliver it throughout the network.Item Applications of Unmanned Vehicles with Wireless Sensor Networks and Surveying(2018-12) Nguyen, An Vn; Becker, Aaron T.; Franchek, Matthew A.; Pan, MiaoAn unmanned vehicle, such as a flying multi-copter, an unmanned rover, a remote-controlled boat, can be used to cover a large area of land, to perform repetitive, tedious yet strenuous tasks for people. We can have an unmanned aerial vehicle (UAV) distribute a network of seismic microphone, used during seismic surveying, in treacherous terrain, free of heavy signal wiring, without risking injury to human workers. A UAV can sweep a large area with a mosquito-zapping net, destructively sampling mosquito population in the area, giving entomology researcher better data about their distribution and behavior through time and space. An unmanned boat, or a UAV, can distribute a drifting wireless sensor network (WSN) into a body of water. The same, or several unmanned vehicles, can then monitor, recharge and finally recollect them. The following thesis presents hardware for all of the above applications, as well as software and algorithms for the unmanned vehicles, and sensor nodes.Item Applying Machine Learning, Asset Allocation, and Risk Management in Selecting Stocks and Building Profitable Portfolio(2022-12-14) Duong, Binh; Nguyen, Hien Van; Chen, Jiefu; Pan, MiaoMany years ago, the stock market became one of the considerable investment channels for many individuals. When the Covid-19 pandemic occurred in the world, many people are placed under full or partial lockdown, and the “stock” keyword became trendy on Google Search. People started discussing investing or trading in the stock market. Approximately, 95% traders lose money due to many factors such as lack of experience, financial knowledge, or fortune. The comparison between investors and traders, which includes all types of traders such as day traders and swing traders, gradually became popular. There are many strategies to invest or trade in the stock market. One of the efficient methods is to select good stocks to build a profitable portfolio. It sounds easy, but it requires many steps and effort including collecting and cleaning data, filtering stocks, gathering potential stocks, predicting stock returns, creating portfolios, filtering portfolios by using risk management, finding weights of each stock by asset allocation, and testing with a custom environment.Item Automating a Seismic Survey Using Heterogenous Sensor Teams and Unmanned Aerial Vehicles(2016-12) Venkata Sudarshan, Srikanth; Becker, Aaron T.; Stewart, Robert R.; Pan, Miao; Han, ZhuSeismic imaging is the primary technique for subsurface exploration. It requires placing a large number of sensors (geophones) in a grid pattern, triggering a seismic event, and recording the propagating waves. The location of hydrocarbons is inferred from these readings. Traditional seismic surveying for hydrocarbons employs human laborers for sensor placement, lays miles of cabling, and then recovers the sensors. Often sites of resource or rescue interest may be difficult or hazardous to access. The major drawbacks of surveying with human deployment are the high costs and time, and risks to humans due to explosives and harsh climatic conditions. Thus, there is a substantial need to automate the process of seismic sensor placement and retrievals using robots. We propose an autonomous, heterogeneous sensor deployment system using UAVs to plant immobile sensors and deploy mobile sensors. Detailed analysis and comparison with traditional surveying were conducted. Hardware experiments and simulations prove the effectiveness of automation regarding cost and time. The proposed system overcame the drawbacks and displayed higher efficiency. The deployed sensors essentially became a wireless sensor network (WSN). Thus traditional batteries cannot sustain a WSN. Energy is the major impediment to the sustainability of WSNs. Most energy is consumed by (i) wireless transmissions of sensed data and (ii) long-distance multi-hop transmissions from the source sensors to the sink. This research also presents an optimal path-planning algorithm for sustaining WSNs and validates the claim with simulations. The research in the future aims at exploring methods to exploit emerging wireless power transfer technology by using UAVs to service the WSNs. These UAVs cut data transmissions from long to short distances by collecting sensed information and replenishing WSN’s energy.Item Big Data Analysis of Complex Networks Using Machine Learning Methods(2016-05) Pan, Erte; Han, Zhu; Ogmen, Haluk; Prasad, Saurabh; Pan, Miao; Li, Husheng; Qian, LijunWith the tremendous development of the modern complex networks such as smart grid and wireless communication domains, the data analysis tasks are significantly involved. In smart grid systems, there are emerging concerns on recognizing energy users' behavior patterns so that the energy trading companies are able to provide customized services. To understand users' usage patterns, efficient grouping methods are required such as clustering or nonparametric Bayesian models in machine learning field. In wireless communication field, a heat topic of locating personal devices and trajectory analysis is drawing more and more attention with the development of advanced personal devices such as the smart phones. Given this background, this dissertation provides a theoretical research in smart grid systems and wireless communications networks with emphases on probabilistic clustering analysis, pricing scheme design, sublinear sampling, tensor voting theory and trajectory pattern recognition. The main contributions of this dissertation include: a comprehensive overview of basic concepts, models and state-of-the-art techniques used in smart grid and trajectory analysis is provided; a novel distance measurement for clustering analysis is proposed from the probabilistic point of view. Moreover, the stopping rules and clustering quality problems have been investigated with proposed novel metrics; the pricing schemes design has been formulated into an optimization problem. The novel sublinear sampling algorithm has been developed to address the computation efficiency in the context of big data; the tensor voting theory has been introduced to the trajectory inference problem and is implemented in the sparse sense to facilitate the computation. The fractal analysis has been employed as a novel method to extract trajectory features for trajectory pattern recognition tasks.Item Big Data Optimization for Distributed Resource Management in Smart Grid(2017-05) Nguyen, Hung Khanh; Han, Zhu; Rajashekara, Kaushik; Pan, Miao; Khodaei, Amin; Mohsenian-Rad, HamedElectric power grids are experiencing the increasing adoption of distributed energy resources, which can bring huge economical and environmental benefit. However, the large-scale penetration of distributed energy resources will make both operations and long-term planning to be more and more complex due to the higher degree of output variability than traditional centralized sources. This variability creates irresistible challenges for grid operators to ensure system security and reliability. In addition, traditional optimization algorithms are no longer applicable for such integrated and complex systems in which economic efficiency, grid reliability, and privacy need to be simultaneously satisfied. Therefore, an innovative optimization framework is critical to tackle the emerging challenges due to the large-scale and independent decision-making nature of distributed resource management problem in the future power system. In this dissertation, we focus on the application of big data optimization methods for distributed resource management problem in smart grid to improve the reliability and security of the distribution system. First, we propose an incentive mechanism design to motivate microgrids to participate in the peak ramp minimization problem for the system to mitigate the ramping effect due to the high penetration of distributed renewable generations. Distributed algorithms to achieve the optimal operation point are proposed, which allow microgrids to execute their computation in either synchronous fashion or asynchronous fashion. Second, a large-scale optimization problem for microgrid optimal scheduling and the load curtailment problem is formulated. We propose a decomposition algorithm and implement parallel computation for the proposed algorithm to run on a computer cluster using the Hadoop MapReduce software framework. Third, a decentralized reactive power compensation model is studied to reduce the power losses and improve the voltage profile for distribution networks. Finally, we consider big data optimization methods for resource allocation problem in wireless network virtualization to prevent traffic disruption against physical network failures.Item Big Data Optimization for Modern Communication Networks(2014-12) Liu, Lanchao; Han, Zhu; Shih, Wei-Chuan; Ogmen, Haluk; Prasad, Saurabh; Pan, Miao; Hong, MingyiThe unprecedented big data in modern communication networks presents us opportunities and challenges. An efficient analytic method for the sheer volume of data is of significant importance for smart grid evolution, intelligent communication network management, efficient medical data management, personalized business model design and smart city development. Meanwhile, the huge volume of data makes it impractical to collect, store and processing in a centralized fashion. Moreover, the massive datasets are noisy, incomplete, heterogeneous, structured, prone to outliers, and vulnerable to cyber-attacks. Overall, we are facing a problem in which the classic resources of computation such as time, space, and energy, are intertwined in complex ways with the massive data sources, and new computational mathematical models as well as methodologies must be explored. With the rapid development of the modern communication networks comes the need of novel algorithms for large-scale data processing and optimization. In this thesis, we investigate the application of big data optimization methods for smart grid security and mobile data traffic management. Firstly, we review the parallel and distributed optimization algorithms based on an alternating direction method of multipliers for solving big data optimization problems. The mathematical backgrounds of the algorithms are given, and the implementations on large-scale computing facilities are also illustrated. Next, the applications of big data processing techniques for smart grid security are studied from two perspectives: how to exploit the inherent structure of the data, and how to deal with the huge size of the data sets. Explored problems are the sparse optimization approach for false data injection detection, and the distributed parallel approach for the security-constrained optimal power flow problem, respectively. Finally, we consider big data optimization methods for data traffic management in mobile cloud computing by two specific application cases: the mobile data offloading in a software defined network at the network edge, and the management of mobile cloud service request allocation and response routing. It is shown by numerical results that effective management and processing of ‘big data’ have the potential to significantly improve smart grid security as well as resource utilization and service quality of the mobile cloud computing.Item Big Data Privacy Preservation for Cyber-Physical Systems(2019-05) Wang, Jingyi; Pan, Miao; Han, Zhu; Prasad, Saurabh; Faghih, Rose T.; Qian, LijunCyber-physical systems (CPS) often referred as ``next generation of engineered systems" are sensing and communication systems that offer tight integration of computation and networking capabilities to monitor and control entities in the physical world. The advent of cloud computing technologies, artificial intelligence and machine learning models has extensively contributed to these multidimensional and complex systems by facilitating a systematic transformation of massive data into information. Though CPS have infiltrated into many areas due to their advantages, big data analytics and privacy are major considerations for building efficient and high-confidence CPS. Many domains of CPS such as smart metering, intelligent transportation, health care, sensor/data aggregation, crowd sensing etc., typically collect huge amounts of data for decision making, where the data may include individual or sensitive information. Since vast amount of information is analyzed, released and calculated by the system to make smart decisions, big data plays a key role as an advanced analysis technique providing more efficient and complete solutions for CPS. However, data privacy breaches during any stage of these large scale systems, either during collection or big data analysis can be an undesirable loss of privacy for the participants and for the entire system. This work focuses on effective big data analytics for CPS and addresses the privacy issues that arise in various CPS applications. Because of their numerous advantages, CPS and its communication networks inevitably become the targets of attackers and malicious users either during data collection, data storage, data transmission, or data processing and computation, keeping users' information at risk. Given these challenges, this work endeavors to develop a series of privacy preserving data analytic and processing methodologies through data driven optimization based on differential privacy; and focuses on effectively integrating the data analysis and data privacy preservation techniques to provide the most desirable solutions for the state-of-the-art CPS with various application-specific requirements.Item Contract Theory Framework for Cryptoeconomics(2022-05-12) Li, Jing; Han, Zhu; Pan, Miao; Shi, Weidong; Nguyen, Hien Van; Niyato, Dusit; Zhang, Xiao-PingCryptoeconomics is the research on how incentives should construct a decentralized and distributed cryptographic system. Economic incentives are used to motivate the efforts and govern the allocation of resources in the cryptoeconomic ecosystem, ensuring specific types of information security qualities. Compared to the costly and time-consuming cryptography, incentives obtained through game theory are much more cost-efficient and easier to implement. However, there lacks sufficient research on the incentive issue of cryptoeconomics. We investigate the various incentives of blockchain networks to fill in the gaps in cryptoeconomics research. The first research focuses on the blockchain network with shards and adopts the security-deposit-based consensus protocol, studying the problem of how to balance the security incentive and the economic incentive. The contract theory is utilized to formulate the problem between temporary blockchain leaders and validators. Compared with fixed deposits, flexible deposits can provide sufficient financial incentives for the participants without losing the security incentives. In the second work, we adopt the cyber insurance idea and propose the insurance contract to help determine the withdrawal delay and the insurance claim to relieve the loss of victims. Specifically, instead of requiring the insurance premium from the validators, the cyber insurer first signs the contract with the blockchain representative (e.g., beacon chain). Then the blockchain representative would sign a series of contracts with the validators. Through the simulations, we demonstrate that the proposed model can provide adaptive insurance contracts for the different validators and keep the profits of the blockchain network and the cyber insurer. In the last work, we propose a random-contract-based scheme to maximize the service provider's revenue and assign the service buyers the feasible service price under the framework of a sidechain linked to the public blockchain. We systematically demonstrate random contracts' superiority under the increasing absolute risk-aversion assumption. The simulation results show that random contracts can provide more significant revenue for sidechains by an average of 24.70% compared to deterministic contracts. Efficient service payments can be reduced by an average of 44.65% compared to the main chain's cost.Item Contract Theory Framework for Wireless Networking(2016-05) Zhang, Yanru; Han, Zhu; Ogmen, Haluk; Prasad, Saurabh; Pan, Miao; Xiong, Zixiang; Qian, LijunWith the rapid development of the modern communication networks, the problem we need to solve is no longer a pure engineering issue. In various heterogeneous network scenarios, there are service providers in need of performing economic analysis on how to ensure third parties' cooperation or attract end-users. In the other way round, third parties or end-users need to evaluate the economic benefits of cooperating or using the services from different service providers. Overall, the current wireless networks are facing a problem in which there is a tight coupling of industry-specific technologies and non-technology related network externality. Contract theory, the 2014 Nobel Prize of economic science, has been widely used in industries, from banking to telecommunications. Particularly, contract theory is an efficient tool in dealing with asymmetric information between employer/seller(s) and employee/buyer(s) by introducing cooperation. In wireless networks, the employer/seller(s) and employee/buyer(s) can be of different roles depending on the scenario under consideration. Thus, there is a great potential to utilize the ideas, methods, and models of contract theory to design efficient wireless network mechanisms. Given this background, this dissertation provides a theoretical research between wireless communications, networking, and economics. Especially, different contract theory models have been applied in various wireless networks scenarios. The main contribution of this dissertation are as follows. An overview of basic concepts, classifications, and models of contract theory is provided. Furthermore, comparisons with existing economics methods in wireless networks are conducted. Applications of contract theory for wireless networks are studied. Specially, three contract theory problems: adverse selection, moral hazard, and a mixed of the two, are applied into device-to-device (D2D) communication, mobile crowdsourcing, cognitive radio network, respectively. Numerical results are provided to show that contract theory can be utilized for developing effective mechanisms for emerging wireless network scenarios such as traffic offloading, mobile crowdsourcing, as well as spectrum trading. The potential and challenges of contract theory as a tool for designing mechanisms in future wireless networks are discussed. This dissertation provides a theoretical research between wireless communications, networking, and economics, in which different contract theory models have been applied in various wireless networks scenarios. This work places a fundamental research on network economics, especially with the framework of contract theory. This research has the potential to contribute to the future of wireless networks network economics area, and have a long term effect on problems such as incentive mechanism and pricing schemes design, resource sharing and trading.Item Control of Magnetic Robots: Solid Medium Transmission and Milli-Scale Magnetic Swimmer(2020-12) Zhao, Haoran; Becker, Aaron T.; Han, Zhu; Faghih, Rose T.; Ruchhoeft, Paul; Pan, Miao; Leclerc, JulienMagnetic robots show great potential for revolutionizing many aspects of medicine and clinical applications. The human body is transparent to a low-frequency magnetic field. Generally, a low frequency is considered less than 300 Hz. Magnetic resonance imaging (MRI) systems typically use a maximum slew rate of 200 mT/m/ms to limit the frequency. MRI is a powerful diagnostic modality for interventions and surgeries. However, MRIs are not used for performing interventions because the MRI has a very high magnetic field and is size constrained. The MRI opening is typically a cylinder that is 30cm in diameter and must accommodate a patient, gradient coils, and the MRI bed. This dissertation provides the design and implementation of a remotely-driven, MR-compatible robotic manipulator, and a force transmission mechanism for controlling that robot. Magnetism is also a promising modality for controlling robots. Magnetically actuated robots could perform minimally invasive surgery. Such robots could be employed for many clinical and biomedical applications, ranging from in vitro to in vivo applications of diagnosis and therapy. Part two of this dissertation examines the control, design optimization, and applications of a spiral shaped magnetic robot. The primary application is focused on blood clot removal. For clot removal, magnetic robots should be controlled and navigated in 3D environments. This requires control algorithms for high accuracy path-following in 3D fluidic environments. The dissertation provides frameworks, design concepts, and control theories for accurate control during blood clot removal. A further change for clot removal is that the clots are removed deep inside the human body. These areas are not visible to cameras, so control of the robots requires imaging techniques. This dissertation presents a process using an ultrasound scanner mounted on a six-axis robot arm to image and tracking the 6 mm long by 2.5 mm diameter magnetic swimmer as it moving in models of human vasculature.Item Delay and Energy Efficient Federated Learning over Heterogeneous Mobile Devices(2023-05-12) Chen, Rui; Pan, Miao; Han, Zhu; Fu, Xin; Gong, Yanmin; Lorenzo, Beatriz; Gnawali, OmprakashThe unprecedented growth in mobile data has created an opportunity for artificial intelligence (AI) to leverage this wealth of information for learning models, which in turn make remarkable improvements on personalized and efficient mobile services, ranging from predictive typing to location-based recommendations. The privacy-sensitive nature of mobile data precludes its central gathering to form the necessary large datasets for the development of high-performance AI models. To overcome this challenge, federated learning (FL) has emerged as a promising distributed machine learning paradigm that enables mobile devices to collaboratively learn a model without sharing raw data. However, practical deployment of FL over mobile devices is very challenging because (i) conventional FL training is power-hungry and time consuming for mobile devices due to interleaved local on device training and communications of model updates, (ii) there are heterogeneous training data across mobile devices, and (iii) mobile devices have hardware heterogeneity in terms of computing and communication capabilities. This dissertation focuses on delay- and energy-efficient FL over mobile devices. In this dissertation, two delay-efficient FL schemes over mobile devices, i.e., SDEFL and FedCR, are proposed to minimize the training latency of mobile devices from the aspects of algorithm design and system design, respectively. Additionally, inspired by our observation of the huge amount of energy saved by high-speed wireless communications during the FL training, we propose a energy-aware local stochastic gradient descent policy to minimize the energy of FL over mobile devices. Furthermore, a full prototype of FL system is implemented to validate the effectiveness of the proposed designs under various FL tasks, learning models, datasets and wireless transmission rates.Item Designing Highly-Efficient Hardware Accelerators for Robust and Automatic Deep Learning Technologies(2022-12-14) Wan, Qiyu; Fu, Xin; Rajashekara, Kaushik; Pan, Miao; Chen, Jinghong; Song, Shuaiwen LeonDeep learning based AI technologies, such as deep convolutional neural networks (DNNs), have recently achieved amazing success in numerous applications, such as image recognition, autonomous driving, and so on. However, there are two critical issues in the conventional DNN applications. The first problem is safety. DNN models can become unreliable due to the uncertainty in data, e.g., insufficient labeled training data, measurement errors and noise in the label. To address this issue, Bayesian deep learning has become an appealing solution since it provides a mathematically grounded framework to quantify uncertainties for model's final prediction. As a key example, Bayesian Neural Networks (BNNs) are one of the most successful Bayesian models being increasingly employed in a wide range of real-world AI applications which demand reliable and robust decisions. However, the nature of BNN stochastic inference and training procedures incurs orders of magnitude higher computational costs than conventional DNN models, which poses a daunting challenge to traditional hardware platforms, such as CPUs/GPUs. The second issue lying in the conventional DNN applications is the laboring-intensive design period. The actual architecture design of a DNN model demands significant amount of efforts and cycles from machine learning experts. Fortunately, the recent emergence of Neural Architecture Search (NAS) has brought the neural architecture design into an era of automation. Nevertheless, the search cost is still prohibitively expensive for practical large-scale deployment in real-world applications. This dissertation focuses on designing high-speed and energy-efficient hardware accelerators for robust and automatic deep learning technologies, i.e. BNN and NAS. In this dissertation, two BNN accelerators, i.e., Fast-BCNN and Shift-BNN, are proposed to accelerate the BNN inference and training, respectively. Furthermore, an efficient in-situ NAS search engine is introduced for large-scale deployment in the real-world applications. The proposed accelerators show promise of solving the challenges during the execution of BNN and NAS workloads efficiently.Item Detecting Cyber-attacks to Smart Grids and Increasing Resiliency Using Data Driven Algorithms(2020-08) Ahmadian, Saeed; Malki, Heidar A.; Han, Zhu; Pan, Miao; Rajashekara, Kaushik; Wang, JianhuiData driven algorithms can be generally divided into two main categories including optimization methods and machine learning approaches. Optimization methods try to find the optimal decision states by finding the feasible boundaries of the problem. On the other hand, machine learning algorithms aim to find the solutions by iterating via small steps toward the optimal answer following the gradient descents. These two data-driven algorithms are widely deployed in many science and engineering fields and in this dissertation, we use both of these methods to address cyber-security issues of smart grids. We first use the optimization algorithm to present two bi-level problems to address the bidding problem in electricity markets and cyber-attack detection in virtual bidding process in electricity markets. We investigate False Data Injection (FDI) problem in smart grids and the approaches the detect attacks. Both models are solved using mathematical programming with equality constraint (MPEC) and the possible cyber-attack's locations and malicious data are identified. We then study the machine learning abilities to learn the cyber-attacker's behavior using real data. We use the Day-ahead (DA) and Real-time (RT) electricity price and demand to create our initial model of the cyber-attacker. Then, we apply a zero-sum game between the cyber-attacker and system defender using novel machine learning method known as Generative Adversarial Networks (GANs). Then, we present a new deep learning structure to model both cyber-attacker and system defender and aslo flexibility of the system defender to learn different possible attacks. We also use another machine learning approach to mitigate the cyber-attacks effects. Particularly, we use Reinforcement Learning (RL) to investigate the optimal possible actions after the cyber-attack happens in the system. In order to model the possible attack's locations we use multi-stage game between the cyber-attacker and system defender. To model the attacker's moves, we use the Hamiltonian Markov Chain Monte Carlo (H-MCMC) and sample from the posterior distribution of the attack's locations. we then train a deep RL network to learn the optimal actions regarding given game stage and possible future game stages.Item Enabling Efficient Neural Network Computation Via Hardware And Software Co-Design(2020-08) Zhang, Xingyao; Fu, Xin; Chen, Jinghong; Pan, Miao; Jackson, David R.; Wu, XuqingIn recent years, the neural networks have achieved great successes in the many area, e.g., automotive driving, medical and Intelligent Personal Assistants (IPAs). Among the neural network models, Long-Short Term Memory network (LSTM) and Capsule Network (CapsNet) are popular but exhibit low efficient when processed on the hardware device. In this dissertation, I introduce two hardware and software co-design approaches to efficiently execute the inference stage of the LSTM and the CapsNet. In the first work, we observe that LSTMs exhibit quite inefficient memory access pattern when executed on mobile GPUs due to the redundant data movements and limited off-chip bandwidth. To address the redundancy, we propose inter-cell level optimizations to improve the data locality across cells with negligible accuracy loss. To relax the pressure on limited offchip memory bandwidth, we propose intra-cell level optimizations that dynamically skip the loads and computations of rows in the weight matrices with trivial contribution to the outputs. We also introduce a light-weighted module to the GPUs architecture for the runtime row skipping in weight matrices. In the second work, CapsNet execution is observed low efficiency due to the execution features of their routing procedure, including massive unshareable intermediate variables and intensive synchronizations. we propose the software-hardware co-designed optimizations, SH-CapsNet, which includes the software-level optimizations named S-CapsNet and a hybrid computing architecture design named PIM-CapsNet . In software-level, S-CapsNet reduces the computation and memory accesses by exploiting the computational redundancy and data similarity of the routing procedure. In hardware-level, the PIM-CapsNet leverages the processing-in-memory capability of today’s 3D stacked memory to conduct the off-chip in-memory acceleration solution for the routing procedure, while pipelining with the GPU’s on-chip computing capability for accelerating CNN types of layers in CapsNet.Item Experiments for the Control and Localization of Robots using Magnetic Induction(2020-12) Soto, Steban S.; Becker, Aaron T.; Pan, Miao; Chen, JiefuMagnetic induction wireless communications are currently being studied as an alternative for short-range communications in underwater robotics and wireless sensor networks. We demonstrate several systems that use magnetic induction coil antennas for robot control and localization. We start with simulations that show a robot executing a controlled approach to a transmitting coil antenna. We describe the performance of our simulations when subjected to external noise. To determine the feasibility of our control and localization algorithms, we perform experiments in air and underwater. These experiments include a study on its design and implementation despite physical constraints.
- «
- 1 (current)
- 2
- 3
- »