You are here
Search results
(1 - 20 of 20)
- Title
- Directed information for complex network analysis from multivariate time series
- Creator
- Liu, Ying
- Date
- 2012
- Collection
- Electronic Theses & Dissertations
- Description
-
Complex networks, ranging from gene regulatory networks in biology to social networks in sociology, havereceived growing attention from the scientific community. The analysis of complex networks employs techniquesfrom graph theory, machine learning and signal processing. In recent years, complex network analysis tools havebeen applied to neuroscience and neuroimaging studies to have a better understanding of the human brain. In thisthesis, we focus on inferring and analyzing the complex...
Show moreComplex networks, ranging from gene regulatory networks in biology to social networks in sociology, havereceived growing attention from the scientific community. The analysis of complex networks employs techniquesfrom graph theory, machine learning and signal processing. In recent years, complex network analysis tools havebeen applied to neuroscience and neuroimaging studies to have a better understanding of the human brain. In thisthesis, we focus on inferring and analyzing the complex functional brain networks underlying multichannelelectroencephalogram (EEG) recordings. Understanding this complex network requires the development of a measureto quantify the relationship between multivariate time series, algorithms to reconstruct the network based on thepairwise relationships, and identification of functional modules within the network.Functional and effective connectivity are two widely studiedapproaches to quantify the connectivity between two recordings.Unlike functional connectivity which only quantifies the statisticaldependencies between two processes by measures such as crosscorrelation, phase synchrony, and mutual information (MI), effectiveconnectivity quantifies the influence one node exerts on anothernode. Directed information (DI) measure is one of the approachesthat has been recently proposed to capture the causal relationshipsbetween two time series. Two major challenges remain with theapplication of DI to multivariate data, which include thecomputational complexity of computing DI with increasing signallength and the accuracy of estimation from limited realizations ofthe data. Expressions that can simplify the computation of theoriginal definition of DI while still quantifying the causalityrelationship are needed. In addition, the advantage of DI overconventionally causality measures such as Granger causality has notbeen fully investigated. In this thesis, we propose time-laggeddirected information and modified directed information to addressthe issue of computational complexity, and compare the performanceof this model free measure with model based measures (e.g. Grangercausality) for different realistic signal models.Once the pairwise DI between two random processes is computed,another problem is to infer the underlying structure of the complexnetwork with minimal false positive detection. We propose to useconditional directed information (CDI) proposed by Kramer to addressthis issue, and introduce the time-lagged conditional directedinformation and modified conditional directed information to lowerthe computational complexity of CDI. Three network inferencealgorithms are presented to infer directed acyclic networks whichcan quantify the causality and also detect the indirect couplingssimultaneously from multivariate data.One last challenge in the study of complex networks, specifically in neuroscience applications, is to identifythe functional modules from multichannel, multiple subject recordings. Most research on community detection inthis area so far has focused on finding the association matrix based on functional connectivity, instead ofeffective connectivity, thus not capturing the causality in the network. In addition, in order to find a modularstructure that best describes all of the subjects in a group, a group analysis strategy is needed. In thisthesis, we propose a multi-subject hierarchical community detection algorithm suitable for a group of weightedand asymmetric (directed) networks representing effective connectivity, and apply the algorithm to multichannelelectroencephalogram (EEG) data.
Show less
- Title
- Kernel methods for biosensing applications
- Creator
- Khan, Hassan Aqeel
- Date
- 2015
- Collection
- Electronic Theses & Dissertations
- Description
-
This thesis examines the design noise robust information retrieval techniques basedon kernel methods. Algorithms are presented for two biosensing applications: (1)High throughput protein arrays and (2) Non-invasive respiratory signal estimation.Our primary objective in protein array design is to maximize the throughput byenabling detection of an extremely large number of protein targets while using aminimal number of receptor spots. This is accomplished by viewing the proteinarray as a...
Show moreThis thesis examines the design noise robust information retrieval techniques basedon kernel methods. Algorithms are presented for two biosensing applications: (1)High throughput protein arrays and (2) Non-invasive respiratory signal estimation.Our primary objective in protein array design is to maximize the throughput byenabling detection of an extremely large number of protein targets while using aminimal number of receptor spots. This is accomplished by viewing the proteinarray as a communication channel and evaluating its information transmission capacity as a function of its receptor probes. In this framework, the channel capacitycan be used as a tool to optimize probe design; the optimal probes being the onesthat maximize capacity. The information capacity is first evaluated for a small scaleprotein array, with only a few protein targets. We believe this is the first effort toevaluate the capacity of a protein array channel. For this purpose models of theproteomic channel's noise characteristics and receptor non-idealities, based on experimental prototypes, are constructed. Kernel methods are employed to extend thecapacity evaluation to larger sized protein arrays that can potentially have thousandsof distinct protein targets. A specially designed kernel which we call the ProteomicKernel is also proposed. This kernel incorporates knowledge about the biophysicsof target and receptor interactions into the cost function employed for evaluation of channel capacity.For respiratory estimation this thesis investigates estimation of breathing-rateand lung-volume using multiple non-invasive sensors under motion artifact and highnoise conditions. A spirometer signal is used as the gold standard for evaluation oferrors. A novel algorithm called the segregated envelope and carrier (SEC) estimation is proposed. This algorithm approximates the spirometer signal by an amplitudemodulated signal and segregates the estimation of the frequency and amplitude in-formation. Results demonstrate that this approach enables effective estimation ofboth breathing rate and lung volume. An adaptive algorithm based on a combination of Gini kernel machines and wavelet filltering is also proposed. This algorithm is titledthe wavelet-adaptive Gini (or WAGini) algorithm, it employs a novel wavelet trans-form based feature extraction frontend to classify the subject's underlying respiratorystate. This information is then employed to select the parameters of the adaptive kernel machine based on the subject's respiratory state. Results demonstrate significantimprovement in breathing rate estimation when compared to traditional respiratoryestimation techniques.
Show less
- Title
- Convolutional neural networks for automated cell detection in magnetic resonance imaging data
- Creator
- Afridi, Muhammad Jamal
- Date
- 2017
- Collection
- Electronic Theses & Dissertations
- Description
-
Cell-based therapy (CBT) is emerging as a promising solution for a large number of serious health issues such as brain injuries and cancer. Recent advances in CBT, has heightened interest in the non-invasive monitoring of transplanted cells in in vivo MRI (Magnetic Resonance Imaging) data. These cells appear as dark spots in MRI scans. However, to date, these spots are manually labeled by experts, which is an extremely tedious and a time consuming process. This limits the ability to conduct...
Show moreCell-based therapy (CBT) is emerging as a promising solution for a large number of serious health issues such as brain injuries and cancer. Recent advances in CBT, has heightened interest in the non-invasive monitoring of transplanted cells in in vivo MRI (Magnetic Resonance Imaging) data. These cells appear as dark spots in MRI scans. However, to date, these spots are manually labeled by experts, which is an extremely tedious and a time consuming process. This limits the ability to conduct large scale spot analysis that is necessary for the long term success of CBT. To address this gap, we develop methods to automate the spot detection task. In this regard we (a) assemble an annotated MRI database for spot detection in MRI; (b) present a superpixel based strategy to extract regions of interest from MRI; (c) design a convolutional neural network (CNN) architecture for automatically characterizing and classifying spots in MRI; (d) propose a transfer learning approach to circumvent the issue of limited training data, and (e) propose a new CNN framework that exploits labeling behavior of the expert in the learning process. Extensive experiments convey the benefits of the proposed methods.
Show less
- Title
- Hardware Algorithms for High-Speed Packet Processing
- Creator
- Norige, Eric
- Date
- 2017
- Collection
- Electronic Theses & Dissertations
- Description
-
The networking industry is facing enormous challenges of scaling devices to support theexponential growth of internet traffic as well as increasing number of features being implemented inside the network. Algorithmic hardware improvements to networking componentshave largely been neglected due to the ease of leveraging increased clock frequency and compute power and the risks of implementing complex hardware designs. As clock frequencyslows its growth, algorithmic solutions become important to...
Show moreThe networking industry is facing enormous challenges of scaling devices to support theexponential growth of internet traffic as well as increasing number of features being implemented inside the network. Algorithmic hardware improvements to networking componentshave largely been neglected due to the ease of leveraging increased clock frequency and compute power and the risks of implementing complex hardware designs. As clock frequencyslows its growth, algorithmic solutions become important to fill the gap between currentgeneration capability and next generation requirements. This paper presents algorithmicsolutions to networking problems in three domains: Deep Packet Inspection(DPI), firewall(and other) ruleset compression and non-cryptographic hashing. The improvements in DPIare two-pronged: first in the area of application-level protocol field extraction, which allowssecurity devices to precisely identify packet fields for targeted validity checks. By usingcounting automata, we achieve precise parsing of non-regular protocols with small, constantper-flow memory requirements, extracting at rates of up to 30gbps on real traffic in softwarewhile using only 112 bytes of state per flow. The second DPI improvement is on the longstanding regular expression matching problem, where we complete the HFA solution to theDFA state explosion problem with efficient construction algorithms and optimized memorylayout for hardware or software implementation. These methods construct automata toocomplex to be constructed by previous methods in seconds, while being capable of 29gbpsthroughput with an ASIC implementation. Firewall ruleset compression enables more firewall entries to be stored in a fixed capacity pattern matching engine, and can also be usedto reorganize a firewall specification for higher performance software matching. A novelrecursive structure called TUF is given to unify the best known solutions to this problemand suggest future avenues of attack. These algorithms, with little tuning, achieve a 13.7%improvement in compression on large, real-life classifiers, and can achieve the same results asexisting algorithms while running 20 times faster. Finally, non-cryptographic hash functionscan be used for anything from hash tables to track network flows to packet sampling fortraffic characterization. We give a novel approach to generating hardware hash functionsin between the extremes of expensive cryptographic hash functions and low quality linearhash functions. To evaluate these mid-range hash functions properly, we develop new evaluation methods to better distinguish non-cryptographic hash function quality. The hashfunctions described in this paper achieve low-latency, wide hashing with good avalanche anduniversality properties at a much lower cost than existing solutions.
Show less
- Title
- Privacy and integrity preserving computation in distributed systems
- Creator
- Chen, Fei
- Date
- 2011
- Collection
- Electronic Theses & Dissertations
- Description
-
Preserving privacy and integrity of private data has become core requirements for many distributed systems across different parties. In these systems, one party may try to compute or aggregate useful information from the private data of other parties. However, this party is not be fully trusted by other parties. Therefore, it is important to design security protocols for preserving such private data. Furthermore, one party may want to query the useful information computed from such private...
Show morePreserving privacy and integrity of private data has become core requirements for many distributed systems across different parties. In these systems, one party may try to compute or aggregate useful information from the private data of other parties. However, this party is not be fully trusted by other parties. Therefore, it is important to design security protocols for preserving such private data. Furthermore, one party may want to query the useful information computed from such private data. However, query results may be modified by a malicious party. Thus, it is important to design query protocols such that query result integrity can be verified.In this dissertation, we study four important privacy and integrity preserving problems for different distributed systems. For two-tiered sensor networks, where storage nodes serve as an intermediate tier between sensors and a sink for storing data and processing queries, we proposed SafeQ, a protocol that prevents compromised storage nodes from gaining information from both sensor collected data and sink issued queries, while it still allows storage nodes to process queries over encrypted data and the sink to detect compromised storage nodes when they misbehave. For cloud computing, where a cloud provider hosts the data of an organization and replies query results to the customers of the organization, we propose novel privacy and integrity preserving schemes for multi-dimensional range queries such that the cloud provider can process encoded queries over encoded data without knowing the actual values, and customers can verify the integrity of query results with high probability. For distributed firewall policies, we proposed the first privacy-preserving protocol for cross-domain firewall policy optimization. For any two adjacent firewalls belonging to two different administrative domains, our protocol can identify in each firewall the rules that can be removed because of the other firewall. For network reachability, one of the key factors for capturing end-to-end network behavior and detecting the violation of security policies, we proposed the first cross-domain privacy-preserving protocol for quantifying network reachability.
Show less
- Title
- Stochastic modeling of routing protocols for cognitive radio networks
- Creator
- Soltani, Soroor
- Date
- 2013
- Collection
- Electronic Theses & Dissertations
- Description
-
Cognitive radios are expected torevolutionize wireless networking because of their ability tosense, manage and share the mobile available spectrum.Efficient utilization of the available spectrum could be significantly improved by incorporating different cognitive radio based networks. Challenges are involved in utilizing the cognitive radios in a network, most of which rise from the dynamic nature of available spectrum that is not present in traditional wireless networks. The set of available...
Show moreCognitive radios are expected torevolutionize wireless networking because of their ability tosense, manage and share the mobile available spectrum.Efficient utilization of the available spectrum could be significantly improved by incorporating different cognitive radio based networks. Challenges are involved in utilizing the cognitive radios in a network, most of which rise from the dynamic nature of available spectrum that is not present in traditional wireless networks. The set of available spectrum blocks(channels) changes randomly with the arrival and departure of the users licensed to a specific spectrum band. These users are known as primary users. If a band is used by aprimary user, the cognitive radio alters its transmission power level ormodulation scheme to change its transmission range and switches to another channel.In traditional wireless networks, a link is stable if it is less prone to interference. In cognitive radio networks, however, a link that is interference free might break due to the arrival of its primary user. Therefore, links' stability forms a stochastic process with OFF and ON states; ON, if the primary user is absent. Evidently, traditional network protocols fail in this environment. New sets of protocols are needed in each layer to cope with the stochastic dynamics of cognitive radio networks.In this dissertation we present a comprehensive stochastic framework and a decision theory based model for the problem of routing packets from a source to a destination in a cognitive radio network. We begin by introducing two probability distributions called ArgMax and ArgMin for probabilistic channel selection mechanisms, routing, and MAC protocols. The ArgMax probability distribution locates the most stable link from a set of available links. Conversely, ArgMin identifies the least stable link. ArgMax and ArgMin together provide valuable information on the diversity of the stability of available links in a spectrum band. Next, considering the stochastic arrival of primary users, we model the transition of packets from one hop to the other by a Semi-Markov process and develop a Primary Spread Aware Routing Protocol (PSARP) that learns the dynamics of the environment and adapts its routing decision accordingly. Further, we use a decision theory framework. A utility function is designed to capture the effect of spectrum measurement, fluctuation of bandwidth availability and path quality. A node cognitively decides its best candidate among its neighbors by utilizing a decision tree. Each branch of the tree is quantified by the utility function and a posterior probability distribution, constructed using ArgMax probability distribution, which predicts the suitability of available neighbors. In DTCR (Decision Tree Cognitive Routing), nodes learn their operational environment and adapt their decision making accordingly. We extend the Decision tree modeling to translate video routing in a dynamic cognitive radio network into a decision theory problem. Then terminal analysis backward induction is used to produce our routing scheme that improves the peak signal-to-noise ratio of the received video.We show through this dissertation that by acknowledging the stochastic property of the cognitive radio networks' environment and constructing strategies using the statistical and mathematical tools that deal with such uncertainties, the utilization of these networks will greatly improve.
Show less
- Title
- High-dimensional learning from random projections of data through regularization and diversification
- Creator
- Aghagolzadeh, Mohammad
- Date
- 2015
- Collection
- Electronic Theses & Dissertations
- Description
-
Random signal measurement, in the form of random projections of signal vectors, extends the traditional point-wise and periodic schemes for signal sampling. In particular, the well-known problem of sensing sparse signals from linear measurements, also known as Compressed Sensing (CS), has promoted the utility of random projections. Meanwhile, many signal processing and learning problems that involve parametric estimation do not consist of sparsity constraints in their original forms. With the...
Show moreRandom signal measurement, in the form of random projections of signal vectors, extends the traditional point-wise and periodic schemes for signal sampling. In particular, the well-known problem of sensing sparse signals from linear measurements, also known as Compressed Sensing (CS), has promoted the utility of random projections. Meanwhile, many signal processing and learning problems that involve parametric estimation do not consist of sparsity constraints in their original forms. With the increasing popularity of random measurements, it is crucial to study the generic estimation performance under the random measurement model. In this thesis, we consider two specific learning problems (named below) and present the following two generic approaches for improving the estimation accuracy: 1) by adding relevant constraints to the parameter vectors and 2) by diversification of the random measurements to achieve fast decaying tail bounds for the empirical risk function.The first problem we consider is Dictionary Learning (DL). Dictionaries are extensions of vector bases that are specifically tailored for sparse signal representation. DL has become increasingly popular for sparse modeling of natural images as well as sound and biological signals, just to name a few. Empirical studies have shown that typical DL algorithms for imaging applications are relatively robust with respect to missing pixels in the training data. However, DL from random projections of data corresponds to an ill-posed problem and is not well-studied. Existing efforts are limited to learning structured dictionaries or dictionaries for structured sparse representations to make the problem tractable. The main motivation for considering this problem is to generate an adaptive framework for CS of signals that are not sparse in the signal domain. In fact, this problem has been referred to as 'blind CS' since the optimal basis is subject to estimation during CS recovery. Our initial approach, similar to some of the existing efforts, involves adding structural constraints on the dictionary to incorporate sparse and autoregressive models. More importantly, our results and analysis reveal that DL from random projections of data, in its unconstrained form, can still be accurate given that measurements satisfy the diversity constraints defined later.The second problem that we consider is high-dimensional signal classification. Prior efforts have shown that projecting high-dimensional and redundant signal vectors onto random low-dimensional subspaces presents an efficient alternative to traditional feature extraction tools such as the principle component analysis. Hence, aside from the CS application, random measurements present an efficient sampling method for learning classifiers, eliminating the need for recording and processing high-dimensional signals while most of the recorded data is discarded during feature extraction. We work with the Support Vector Machine (SVM) classifiers that are learned in the high-dimensional ambient signal space using random projections of the training data. Our results indicate that the classifier accuracy can be significantly improved by diversification of the random measurements.
Show less
- Title
- Unconstrained 3D face reconstruction from photo collections
- Creator
- Roth, Joseph
- Date
- 2016
- Collection
- Electronic Theses & Dissertations
- Description
-
This thesis presents a novel approach for 3D face reconstruction from unconstrained photo collections. An unconstrained photo collection is a set of face images captured under an unknown and diverse variation of poses, expressions, and illuminations. The output of the proposed algorithm is a true 3D face surface model represented as a watertight triangulated surface with albedo data colloquially referred to as texture information. Reconstructing a 3D understanding of a face based on 2D input...
Show moreThis thesis presents a novel approach for 3D face reconstruction from unconstrained photo collections. An unconstrained photo collection is a set of face images captured under an unknown and diverse variation of poses, expressions, and illuminations. The output of the proposed algorithm is a true 3D face surface model represented as a watertight triangulated surface with albedo data colloquially referred to as texture information. Reconstructing a 3D understanding of a face based on 2D input is a long-standing computer vision problem. Traditional photometric stereo-based reconstruction techniques work on aligned 2D images and produce a 2.5D depth map reconstruction. We extend face reconstruction to work with a true 3D model, allowing us to enjoy the benefits of using images from all poses, up to and including profiles. To use a 3D model, we propose a novel normal field-based Laplace editing technique which allows us to deform a triangulated mesh to match the observed surface normals. Unlike prior work that require large photo collections, we formulate an approach to adapt to photo collections with few images of potentially poor quality. We achieve this through incorporating prior knowledge about face shape by fitting a 3D Morphable Model to form a personalized template before using a novel analysis-by-synthesis photometric stereo formulation to complete the fine face details. A structural similarity-based quality measure allows evaluation in the absence of ground truth 3D scans. Superior large-scale experimental results are reported on Internet, synthetic, and personal photo collections.
Show less
- Title
- Efficient and secure system design in wireless communications
- Creator
- Song, Tianlong
- Date
- 2016
- Collection
- Electronic Theses & Dissertations
- Description
-
Efficient and secure information transmission lies in the core part of wireless system design and networking. Comparing with its wired counterpart, in wireless communications, the total available spectrum has to be shared by different services. Moreover, wireless transmission is more vulnerable to unauthorized detection, eavesdropping and hostile jamming due to the lack of a protective physical boundary.Today, the two most representative highly efficient communication systems are CDMA (used...
Show moreEfficient and secure information transmission lies in the core part of wireless system design and networking. Comparing with its wired counterpart, in wireless communications, the total available spectrum has to be shared by different services. Moreover, wireless transmission is more vulnerable to unauthorized detection, eavesdropping and hostile jamming due to the lack of a protective physical boundary.Today, the two most representative highly efficient communication systems are CDMA (used in 3G) and OFDM (used in 4G), and OFDM is regarded as the most efficient system. This dissertation will focus on two topics: (1) Explore more spectrally efficient system design based on the 4G OFDM scheme; (2) Investigate robust wireless system design and conduct capacity analysis under different jamming scenarios. The main results are outlined as follows.First, we develop two spectrally efficient OFDM-based multi-carrier transmission schemes: one with message-driven idle subcarriers (MC-MDIS), and the other with message-driven strengthened subcarriers (MC-MDSS). The basic idea in MC-MDIS is to carry part of the information, named carrier bits, through idle subcarrier selection while transmitting the ordinary bits regularly on all the other subcarriers. When the number of subcarriers is much larger than the adopted constellation size, higher spectral and power efficiency can be achieved comparing with OFDM. In MC-MDSS, the idle subcarriers are replaced by strengthened ones, which, unlike idle ones, can carry both carrier bits and ordinary bits. Therefore, MC-MDSS achieves even higher spectral efficiency than MC-MDIS.Second, we consider jamming-resistant OFDM system design under full-band disguised jamming, where the jamming symbols are taken from the same constellation as the information symbols over each subcarrier. It is shown that due to the symmetricity between the authorized signal and jamming, the BER of the traditional OFDM system is lower bounded by a modulation specific constant. We develop an optimal precoding scheme, which minimizes the BER of OFDM systems under full-band disguised jamming. It is shown that the most efficient way to combat full-band disguised jamming is to concentrate the total available power and distribute it uniformly over a particular number of subcarriers instead of the entire spectrum. The precoding scheme is further randomized to reinforce the system jamming resistance.Third, we consider jamming mitigation for CDMA systems under disguised jamming, where the jammer generates a fake signal using the same spreading code, constellation and pulse shaping filter as that of the authorized signal. Again, due to the symmetricity between the authorized signal and jamming, the receiver cannot really distinguish the authorized signal from jamming, leading to complete communication failure. In this research, instead of using conventional scrambling codes, we apply advanced encryption standard (AES) to generate the security-enhanced scrambling codes. Theoretical analysis shows that: the capacity of conventional CDMA systems without secure scrambling under disguised jamming is actually zero, while the capacity can be significantly increased by secure scrambling.Finally, we consider a game between a power-limited authorized user and a power-limited jammer, who operate independently over the same spectrum consisting of multiple bands. The strategic decision-making is modeled as a two-party zero-sum game, where the payoff function is the capacity that can be achieved by the authorized user in presence of the jammer. We first investigate the game under AWGN channels. It is found that: either for the authorized user to maximize its capacity, or for the jammer to minimize the capacity of the authorized user, the best strategy is to distribute the power uniformly over all the available spectrum. Then, we consider fading channels. We characterize the dynamic relationship between the optimal signal power allocation and the optimal jamming power allocation, and propose an efficient two-step water pouring algorithm to calculate them.
Show less
- Title
- Harnessing Low-Pass Filter Defects for Improving Wireless Link Performance : Measurements and Applications
- Creator
- Ameli Renani, Alireza
- Date
- 2018
- Collection
- Electronic Theses & Dissertations
- Description
-
The design trade-offs of transceiver hardware are crucial to the performance of wireless systems. The effect of such trade-offs on individual analog and digital components are vigorously studied, but their systemic impacts beyond component-level remain largely unexplored. In this dissertation, we present an in-depth study to characterize the surprisingly notable systemic impacts of low-pass filter design, which is a small yet indispensable component used for shaping spectrum and rejecting...
Show moreThe design trade-offs of transceiver hardware are crucial to the performance of wireless systems. The effect of such trade-offs on individual analog and digital components are vigorously studied, but their systemic impacts beyond component-level remain largely unexplored. In this dissertation, we present an in-depth study to characterize the surprisingly notable systemic impacts of low-pass filter design, which is a small yet indispensable component used for shaping spectrum and rejecting interference. Using a bottom-up approach, we examine how signal-level distortions caused by the trade-offs of low-pass filter design propagate to the upper-layers of wireless communication, reshaping bit error patterns and degrading link performance of today's 802.11 systems. Moreover, we propose a novel unequal error protection algorithm that harnesses low-pass filter defects for improving wireless LAN throughput, particularly to be used in forward error correction, channel coding, and applications such as video streaming. Lastly, we conduct experiments to evaluate the unequal error protection algorithm in video streaming, and we present substantial enhancements of video quality in mobile environments.
Show less
- Title
- Scalable pulsed mode computation architecture using integrate and fire structure based on margin propagation
- Creator
- Hindo, Thamira
- Date
- 2014
- Collection
- Electronic Theses & Dissertations
- Description
-
Neuromorphic computing architectures mimic the brain to implement efficient computations for sensory applications in a different way from that of the traditional Von Neumann architecture. The goal of neuromorphic computing systems is to implement sensory devices and systems that operate as efficiently as their biological equivalents. Neuromorphic computing consists of several potential components including parallel processing instead of synchronous processing, hybrid (pulse) computation...
Show moreNeuromorphic computing architectures mimic the brain to implement efficient computations for sensory applications in a different way from that of the traditional Von Neumann architecture. The goal of neuromorphic computing systems is to implement sensory devices and systems that operate as efficiently as their biological equivalents. Neuromorphic computing consists of several potential components including parallel processing instead of synchronous processing, hybrid (pulse) computation instead of digital computation, neuron models as a basic core of the processing instead of the arithmetic logic units, and analog VLSI design instead of digital VLSI design. In this work a new neuromorphic computing architecture is proposed and investigated for the implementation of algorithms based on using the pulsed mode with a neuron-based circuit.The proposed architecture goal is to implement approximate non-linear functions that are important components of signal processing algorithms. Some of the most important signal processing algorithms are those that mimic biological systems such as hearing, sight and touch. The designed architecture is pulse mode and it maps the functions into an algorithm called margin propagation. The designed structure is a special network of integrate-and-fire neuron-based circuits that implement the margin propagation algorithm using integration and threshold operations embedded in the transfer function of the neuron model. The integrate-and-fire neuron units in the network are connected together through excitatory and inhibitory paths to impose constraints on the network firing-rate. The advantages of the pulse-based, integrate-and-fire margin propagation (IFMP) algorithmic unit are to implement complex non-linear and dynamic programming functions in a scalable way; to implement functions using cascaded design in parallel or serial architecture; to implement the modules in low power and small size circuits of analog VLSI; and to achieve a wide dynamic range since the input parameters of IFMP module are mapped in the logarithmic domain.The newly proposed IFMP algorithmic unit is investigated both on a theoretically basis and an experimental performance basis. The IFMP algorithmic unit is implemented with a low power analog circuit. The circuit is simulated using computer aided design tools and it is fabricated in a 0.5 micron CMOS process. The hardware performance of the fabricated IFMP algorithmic architecture is also measured. The application of the IFMP algorithmic architecture is investigate for three signal processing algorithms including sequence recognition, trace recognition using hidden Markov model and binary classification using a support vector machine. Additionally, the IFMP architecture is investigated for the application of the winner-take-all algorithm, which is important for hearing, sight and touch sensor systems.
Show less
- Title
- A multivariate time-frequency based phase synchrony measure and applications to dynamic brain network analysis
- Creator
- Mutlu, Ali Yener
- Date
- 2012
- Collection
- Electronic Theses & Dissertations
- Description
-
Irregular, non-stationary, and noisy multichannel data are abound in many fields of research. Observations of multichannel data in nature include changes in weather, the dynamics of satellites in the solar system, the time evolution of the magnetic field of celestial bodies, population growth in ecology and the dynamics of the action potentials in neurons.One particular application of interest is the functional integration of neuronal networks in the human brain. Human brain is known to be...
Show moreIrregular, non-stationary, and noisy multichannel data are abound in many fields of research. Observations of multichannel data in nature include changes in weather, the dynamics of satellites in the solar system, the time evolution of the magnetic field of celestial bodies, population growth in ecology and the dynamics of the action potentials in neurons.One particular application of interest is the functional integration of neuronal networks in the human brain. Human brain is known to be one of the most complex biological systems and quantifying functional neural coordination in the brain is a fundamental problem. It has been recently proposed that networks of highly nonlinear and non-stationary reciprocal interactions are the key features of functional integration. Among many linear and nonlinear measures of dependency, time-varying phase synchrony has been proposed as a promising measure of connectivity. Current state-of-the-art in time-varying phase estimation uses either the Hilbert transform or the complex wavelet transform of the signals. Both of these methods have some major drawbacks such as the assumption that the signals are narrowband for the Hilbert transform and the non-uniform time-frequency resolution inherent to the wavelet analysis. Furthermore, the current phase synchrony measures are limited to quantifying bivariate relationships and do not reveal any information about multivariate synchronization patterns which are important for understanding the underlying oscillatory networks.In this dissertation, a new phase estimation method based on the Rihaczek distribution and Reduced Interference Rihaczek distribution belonging to Cohen's class is proposed. These distributions offer phase estimates with uniformly high time-frequency resolution which can be used for defining time and frequency dependent phase synchrony within the same frequency band as well as across different frequency bands. Properties of the phase estimator and the corresponding phase synchrony measure are evaluated both analytically and through simulations showing the effectiveness of the new measures compared to existing ones. The proposed distribution is then extended to quantify the cross-frequency phase synchronization between two signals across different frequencies. In addition, a cross frequency-spectral lag distribution is introducedto quantify the amount of amplitude modulation between signals. Furthermore, the notion of bivariate synchrony is extended to multivariate synchronization to quantify the relationships within and across groups of signals. Measures of multiple correlation and complexity are used as well as a more direct multivariate synchronization measure, `Hyperspherical Phase Synchrony', is proposed. This new measure is based on computing pairwise phase differences to create a multidimensional phase difference vector and mapping this vector to a high dimensional space. Hyperspherical phase synchrony offers lower computational complexity and is more robust to noise compared to the existing measures.Finally, a subspace analysis framework is proposed for studying time-varying evolution of functional brain connectivity. The proposed approach identifies event intervals accounting for the underlying neurophysiological events and extracts key graphs for describing the particular intervals with minimal redundancy. Results from the application to EEG data indicate the effectiveness of the proposed framework in determining the event intervals and summarizing brain activity with a few number of representative graphs.
Show less
- Title
- Theory, synthesis and implementation of current-mode CMOS piecewise-linear circuits using margin propagation
- Creator
- Gu, Ming
- Date
- 2012
- Collection
- Electronic Theses & Dissertations
- Description
-
Achieving high energy-efficiency is a key requirement for many emerging smart sensors and portable computing systems. While digital signal processing (DSP) has been the de-facto technique for implementing ultra-low power systems, analog signal processing (ASP) provides an attractive and alternate approach that can not only achieve high energy efficiency but also high computational density. Conventional ASP techniques are based on a top-down design approach, where proven mathematical...
Show moreAchieving high energy-efficiency is a key requirement for many emerging smart sensors and portable computing systems. While digital signal processing (DSP) has been the de-facto technique for implementing ultra-low power systems, analog signal processing (ASP) provides an attractive and alternate approach that can not only achieve high energy efficiency but also high computational density. Conventional ASP techniques are based on a top-down design approach, where proven mathematical principles and related algorithms are mapped and emulated using computational primitives inherent in the device physics. An example being the translinear principle, which is the state-of-the-art ASP technique, that uses the exponential current-to-voltage characteristics for designing ultra-low-power analog processors. However, elegant formulations could result from a bottom-up approach where device and bias independent computational primitives (e.g. current and charge conservation principles) are used for designing "approximate" analog signal processors. The hypothesis of this proposal is that many signal processing algorithms exhibit an inherent calibration ability due to which their performance remains unaffected by the use of "approximate" analog computing techniques. In this research, we investigate the theory, synthesis and implementation of high performance analog processors using a novel piecewise-linear (PWL) approximation algorithm called margin propagation (MP). MP principle utilizes only basic conservation laws of physical quantities (current, charge, mass, energy) for computing and therefore is scalable across devices (silicon, MEMS, microfluidics). However, there are additional advantages of MP-based processors when implemented using CMOS current-mode circuits, which includes: 1) the operation of the MP processor requires only addition, subtraction and threshold operations and hence is independent of transistor biasing (weak, moderate and strong inversion) and robust to variations in environmental conditions (e.g. temperature); and 2) improved dynamic range and faster convergence as compared to the translinear implementations. We verify our hypothesis using two analog signal processing applications: (a) design of high-performance analog low-density parity check (LDPC) decoders for applications in sensor networks; and (b) design of ultra-low-power analog support vector machines (SVM) for smart sensors. Our results demonstrate that an algorithmic framework for designing margin propagation (MP) based LDPC decoders can be used to trade-off its BER performance with its energy efficiency, making the design attractive for applications with adaptive energy-BER constraints. We have verified this trade-off using an analog current-mode implementation of an MP-based (32,8) LDPC decoder. Measured results from prototypes fabricated in a 0.5 μm CMOS process show that the BER performance of the MP-based decoder outperforms a benchmark state-of-the-art min-sum decoder at SNR levels greater than 3.5 dB and can achieve energy efficiencies greater than 100pJ/bit at a throughput of 12.8 Mbps. In the second part of this study, MP principle is used for designing an energy-scalable support vector machine (SVM) whose power and speed requirements can be configured dynamically without any degradation in performance. We have verified the energy-scaling property using a current-mode implementation of an SVM operating with 8 dimensional feature vectors and 18 support vectors. The prototype fabricated in a 0.5μm CMOS process has integrated an array of floating gate transistors that serve as storage for up to 2052 SVM parameters. The SVM prototype also integrates novel circuits that have been designed for interfacing with an external digital processor. This includes a novel current-input current-output logarithmic amplifier circuit that can achieve a dynamic range of 120dB while consuming nanowatts of power. Another novel circuit includes a varactor based temperature compensated floating-gate memory that demonstrates a superior programming range than other temperature compensated floating-gate memories.
Show less
- Title
- Biometric template security
- Creator
- Nagar, Abhishek
- Date
- 2012
- Collection
- Electronic Theses & Dissertations
- Description
-
"This dissertation provides a thorough analysis of the vulnerabilities of a biometric recognition system with emphasis on the vulnerabilities related to the information stored in biometric systems in the form of biometric templates."--From abstract.
- Title
- Exploiting node mobility for energy optimization in wireless sensor networks
- Creator
- El-Moukaddem, Fatme Mohammad
- Date
- 2012
- Collection
- Electronic Theses & Dissertations
- Description
-
Wireless Sensor Networks (WSNs) have become increasingly available for data-intensive applications such as micro-climate monitoring, precision agriculture, and audio/video surveillance. A key challenge faced by data-intensive WSNs is to transmit the sheer amount of data generated within an application's lifetime to the base station despite the fact that sensor nodes have limited power supplies such as batteries or small solar panels. The availability of numerous low-cost robotic units (e.g....
Show moreWireless Sensor Networks (WSNs) have become increasingly available for data-intensive applications such as micro-climate monitoring, precision agriculture, and audio/video surveillance. A key challenge faced by data-intensive WSNs is to transmit the sheer amount of data generated within an application's lifetime to the base station despite the fact that sensor nodes have limited power supplies such as batteries or small solar panels. The availability of numerous low-cost robotic units (e.g. Robomote and Khepera) has made it possible to construct sensor networks consisting of mobile sensor nodes. It has been shown that the controlled mobility offered by mobile sensors can be exploited to improve the energy efficiency of a network.In this thesis, we propose schemes that use mobile sensor nodes to reduce the energy consumption of data-intensive WSNs. Our approaches differ from previous work in two main aspects. First, our approaches do not require complex motion planning of mobile nodes, and hence can be implemented on a number of low-cost mobile sensor platforms. Second, we integrate the energy consumption due to both mobility and wireless communications into a holistic optimization framework.We consider three problems arising from the limited energy in the sensor nodes. In the first problem, the network consists of mostly static nodes and contains only a few mobile nodes. In the second and third problems, we assume essentially that all nodes in the WSN are mobile. We first study a new problem called max-data mobile relay configuration (MMRC) that finds the positions of a set of mobile sensors, referred to as relays, that maximize the total amount of data gathered by the network during its lifetime. We show that the MMRC problem is surprisingly complex even for a trivial network topology due to the joint consideration of the energy consumption of both wireless communication and mechanical locomotion. We present optimal MMRC algorithms and practical distributed implementations for several important network topologies and applications. Second, we consider the problem of minimizing the total energy consumption of a network. We design an iterative algorithm that improves a given configuration by relocating nodes to new positions. We show that this algorithm converges to the optimal configuration for the given transmission routes. Moreover, we propose an efficient distributed implementation that does not require explicit synchronization. Finally, we consider the problem of maximizing the lifetime of the network. We propose an approach that exploits the mobility of the nodes to balance the energy consumption throughout the network. We develop efficient algorithms for single and multiple round approaches. For all three problems, we evaluate the efficiency of our algorithms through simulations. Our simulation results based on realistic energy models obtained from existing mobile and static sensor platforms show that our approaches significantly improve the network's performance and outperform existing approaches.
Show less
- Title
- Graph-based methods for inferring neuronal connectivity from spike train ensembles
- Creator
- El-dawlatly, Seif El-din
- Date
- 2011
- Collection
- Electronic Theses & Dissertations
- Description
-
Understanding the brain's inner workings requires studying the underlying complex networks that bind its basic computational elements, the neurons. Advances in extracellular neural recording techniques have enabled simultaneous recording of spike trains from multiple single neurons in awake, behaving subjects. Yet, devising methods to infer connectivity among these neurons has been significantly lacking. We introduce a connectivity inference framework based on graphical models. We first infer...
Show moreUnderstanding the brain's inner workings requires studying the underlying complex networks that bind its basic computational elements, the neurons. Advances in extracellular neural recording techniques have enabled simultaneous recording of spike trains from multiple single neurons in awake, behaving subjects. Yet, devising methods to infer connectivity among these neurons has been significantly lacking. We introduce a connectivity inference framework based on graphical models. We first infer the functional connectivity between neurons by searching for clusters of statistically dependent spike trains. We then infer the effective connectivity between neurons within each cluster by building Dynamic Bayesian Network (DBN) model fit to the spike train data. Using probabilistic models of neuronal firing, we demonstrate the utility of this framework to infer neuronal connectivity in moderate and large scale networks with a substantial gain in performance compared to classical methods. We further use this framework to examine the role of spike timing correlation in infragranular layer V of the primary somatosensory cortex (S1) of the rat during unilateral whisker stimulation in vivo. Stable, whisker-specific networks provided more information about the stimulus than individual neurons' response. We finally demonstrate how this framework enables tracking and quantifying plastic changes in connectivity in biologically-plausible models of spike-timing-dependent-plasticity as well as changes in S1 response maps following sensory deprivation in the awake, behaving rat.
Show less
- Title
- Implantable VLSI systems for compression and communication in wireless biosensor recording arrays
- Creator
- Kamboh, Awais Mehmood
- Date
- 2010
- Collection
- Electronic Theses & Dissertations
- Description
-
Successful use of microelectrode arrays to record neural activity in the cortex has opened new opportunities for scientists to decode the intricate functionality of the human brain and the behavior of neurons that enable its complex operation. The resulting brain-machine interface devices play a critical role in enabling patients with neural disorders to achieve a better lifestyle. Such interfaces provide a direct interface to the brain and show great promise in many biomedical applications...
Show moreSuccessful use of microelectrode arrays to record neural activity in the cortex has opened new opportunities for scientists to decode the intricate functionality of the human brain and the behavior of neurons that enable its complex operation. The resulting brain-machine interface devices play a critical role in enabling patients with neural disorders to achieve a better lifestyle. Such interfaces provide a direct interface to the brain and show great promise in many biomedical applications.This thesis explores some of the major obstacles impeding the advance of wireless neural implants and addresses them through development of highly efficient algorithms and implantable hardware. An overwhelming amount of data is generated by the microelectrode arrays, resulting in a data bandwidth bottleneck. To overcome this problem, an implantable system has been devised to enable control over the amount of data that must be transmitted without compromising the information contained in the array of neural signals. Furthermore, the nature of the wireless communication channel across the skin tissue is not well characterized. In this thesis, solutions have been developed to maximize that data throughput and enable unfailing yet low-power communication of bidirectional data between the implanted device and the external world. Finally, a unified energy-efficient, implantable CMOS integrated circuit was developed to address these two critical problems. The resulting integrated solution ensures seamless multi-modal operation, and thus establishes a pathway to the design of next-generation neuroprosthetics devices. Although the motivation for this thesis comes from the field of neuroprosthetics, the solutions devised are pertinent to a wide range of implantable applications.
Show less
- Title
- Robust signal processing methods for miniature acoustic sensing, separation, and recognition
- Creator
- Fazel, Amin
- Date
- 2012
- Collection
- Electronic Theses & Dissertations
- Description
-
One of several emerging areas where micro-scale integration promises significant breakthroughs is in the field of acoustic sensing. However, separation, localization, and recognition of acoustic sources using micro-scale microphone arrays poses a significant challenge due to fundamental limitations imposed by the physics of sound propagation. The smaller the distance between the recording elements, the more difficult it is to measure localization and separation cues and hence it is more...
Show moreOne of several emerging areas where micro-scale integration promises significant breakthroughs is in the field of acoustic sensing. However, separation, localization, and recognition of acoustic sources using micro-scale microphone arrays poses a significant challenge due to fundamental limitations imposed by the physics of sound propagation. The smaller the distance between the recording elements, the more difficult it is to measure localization and separation cues and hence it is more difficult to recognize the acoustic sources of interest. The objective of this research is to investigate signal processing and machine learning techniques that can be used for noise-robust acoustic target recognition using miniature microphone arrays.The first part of this research focuses on designing "smart" analog-to-digital conversion (ADC) algorithms that can enhance acoustic cues in sub-wavelength microphone arrays. Many source separation algorithms fail to deliver robust performance when applied to signals recorded using high-density sensor arrays where the distance between sensor elements is much less than the wavelength of the signals. This can be attributed to limited dynamic range (determined by analog-to-digital conversion) of the sensor which is insufficientto overcome the artifacts due to large cross-channel redundancy, non-homogeneous mixing and high-dimensionality of the signal space. We propose a novel framework that overcomes these limitations by integrating statistical learning directly with the signal measurement (analog-to-digital) process which enables high fidelity separation of linear instantaneous mixture. At the core of the proposed ADC approach is a min-max optimization of a regularized objective function that yields a sequence of quantized parameters which asymptotically tracks the statistics of the input signal. Experiments with synthetic and real recordings demonstrate consistent performance improvements when the proposed approach is used as the analog-to-digital front-end to conventional source separation algorithms.The second part of this research focuses on investigating a novel speech feature extraction algorithm that can recognize auditory targets (keywords and speakers) using noisy recordings. The features known as Sparse Auditory Reproducing Kernel (SPARK) coefficients are extracted under the hypothesis that the noise-robust information in speech signal is embedded in a subspace spanned by sparse, regularized, over-complete, non-linear, and phase-shifted gammatone basis functions. The feature extraction algorithm involves computing kernel functions between the speech data and pre-computed set of phased-shifted gammatone functions, followed by a simple pooling technique ("MAX" operation). In this work, we present experimental results for a hidden Markov model (HMM) based speech recognition system whose performance has been evaluated on a standard AURORA 2 dataset. The results demonstrate that the SPARK features deliver significant and consistent improvements in recognition accuracy over the standard ETSI STQ WI007 DSR benchmark features. We have also verified the noise-robustness of the SPARK features for the task of speaker verification. Experimental results based on the NIST SRE 2003 dataset show significant improvements when compared to a standard Mel-frequency cepstral coefficients (MFCCs) based benchmark.
Show less
- Title
- MEASURING AND MODELING THE EFFECTS OF SEA LEVEL RISE ON NEAR-COASTAL RIVERINE REGIONS : A GEOSPATIAL COMPARISON OF THE SHATT AL-ARAB RIVER IN SOUTHERN IRAQ WITH THE MISSISSIPPI RIVER DELTA IN SOUTHERN LOUISIANA, USA.
- Creator
- Kadhim, Ameen Awad
- Date
- 2018
- Collection
- Electronic Theses & Dissertations
- Description
-
There is a growing debate among scientists on how sea level rise (SLR) will impact coastal environments, particularly in countries where economic activities are sustained along these coasts. An important factor in this debate is how best to characterize coastal environmental impacts over time. This study investigates the measurement and modeling of SLR and effects on near-coastal riverine regions. The study uses a variety of data sources, including satellite imagery from 1975 to 2017, digital...
Show moreThere is a growing debate among scientists on how sea level rise (SLR) will impact coastal environments, particularly in countries where economic activities are sustained along these coasts. An important factor in this debate is how best to characterize coastal environmental impacts over time. This study investigates the measurement and modeling of SLR and effects on near-coastal riverine regions. The study uses a variety of data sources, including satellite imagery from 1975 to 2017, digital elevation data and previous studies. This research is focusing on two of these important regions: southern Iraq along the Shatt Al-Arab River (SAR) and the southern United States in Louisiana along the Mississippi River Delta (MRD). These sites are important for both their extensive low-lying land and for their significant coastal economic activities. The dissertation consists of six chapters. Chapter one introduces the topic. Chapter two compares and contrasts bothregions and evaluates escalating SLR risk. Chapter three develops a coupled human and natural system (CHANS) perspective for SARR to reveal multiple sources of environmental degradation in this region. Alfa century ago SARR was an important and productive region in Iraq that produced fruits like dates, crops, vegetables, and fish. By 1975 the environment of this region began to deteriorate, and since then, it is well-documented that SARR has suffered under human and natural problems. In this chapter, I use the CHANS perspective to identify the problems, and which ones (human or natural systems) are especially responsible for environmental degradation in SARR. I use several measures of ecological, economic, and social systems to outline the problems identified through the CHANS framework. SARR has experienced extreme weather changes from 1975 to 2017 resulting in lower precipitation (-17mm) and humidity (-5.6%), higher temperatures (1.6 C), and sea level rise, which are affecting the salinity of groundwater and Shatt Al Arab river water. At the same time, human systems in SARR experienced many problems including eight years of war between Iraq and Iran, the first Gulf War, UN Security Council imposed sanctions against Iraq, and the second Gulf War. I modeled and analyzed the regions land cover between 1975 and 2017 to understand how the environment has been affected, and found that climate change is responsible for what happened in this region based on other factors. Chapter four constructs and applies an error propagation model to elevation data in the Mississippi River Delta region (MRDR). This modeling both reduces and accounts for the effects of digital elevation model (DEM) error on a bathtub inundation model used to predict the SLR risk in the region. Digital elevation data is essential to estimate coastal vulnerability to flooding due to sea level rise. Shuttle Radar Topography Mission (SRTM) 1 Arc-Second Global is considered the best free global digital elevation data available. However, inundation estimates from SRTM are subject to uncertainty due to inaccuracies in the elevation data. Small systematic errors in low, flat areas can generate large errors in inundation models, and SRTM is subject to positive bias in the presence of vegetation canopy, such as along channels and within marshes. In this study, I conduct an error assessment and develop statistical error modeling for SRTM to improve the quality of elevation data in these at-risk regions. Chapter five applies MRDR-based model from chapter four to enhance the SRTM 1 Arc-Second Global DEM data in SARR. As such, it is the first study to account for data uncertainty in the evaluation of SLR risk in this sensitive region. This study transfers an error propagation model from MRDR to the Shatt al-Arab river region to understand the impact of DEM error on an inundation model in this sensitive region. The error propagation model involves three stages. First, a multiple regression model, parameterized from MRDR, is used to generate an expected DEM error surface for SARR. This surface is subtracted from the SRTM DEM for SARR to adjust it. Second, residuals from this model are simulated for SARR: these are mean-zero and spatially autocorrelated with a Gaussian covariance model matching that observed in MRDR by convolution filtering of random noise. More than 50 realizations of error were simulated to make sure a stable result was realized. These realizations were subtracted from the adjusted SRTM to produce DEM realizations capturing potential variation. Third, the DEM realizations are each used in bathtub modeling to estimate flooding area in the region with 1 m of sea level rise. The distribution of flooding estimates shows the impact of DEM error on uncertainty in inundation likelihood, and on the magnitude of total flooding. Using the adjusted DEM realizations 47 ± 2 percent of the region is predicted to flood, while using the raw SRTM DEM only 28% of the region is predicted to flood.
Show less
- Title
- Smartphone-Based Sensing Systems for Data-Intensive Applications
- Creator
- Moazzami, Mohammad-Mahdi
- Date
- 2017
- Collection
- Electronic Theses & Dissertations
- Description
-
Supported by advanced sensing capabilities, increasing computational resources and the advances in Artificial Intelligence, smartphones have become our virtual companions in our daily life. An average modern smartphone is capable of handling a wide range of tasks including navigation, advanced image processing, speech processing, cross app data processing and etc. The key facet that is common in all of these applications is the data intensive computation. In this dissertation we have taken...
Show moreSupported by advanced sensing capabilities, increasing computational resources and the advances in Artificial Intelligence, smartphones have become our virtual companions in our daily life. An average modern smartphone is capable of handling a wide range of tasks including navigation, advanced image processing, speech processing, cross app data processing and etc. The key facet that is common in all of these applications is the data intensive computation. In this dissertation we have taken steps towards the realization of the vision that makes the smartphone truly a platform for data intensive computations by proposing frameworks, application and algorithmic solutions. We followed a data-driven approach to the system design. To this end, several challenges must be addressed before smartphones can be used as a system platform for data-intensive applications. The major challenge addressed in this dissertation include high power consumption, high computation cost in advance machine learning algorithms, lack of real-time functionalities, lack of embedded programming support, heterogeneity in the apps, communication interfaces and programming abstractions and lack of customized data processing libraries. The contribution of this dissertation can be summarized as follows. We presented the design, implementation and evaluation of the ORBIT framework, which represents the first system that combines the design requirements of a machine learning system and sensing system together at the same time. We ported for the first time off-the-shelf machine learning algorithms for real-time sensor data processing to smartphone devices. In this process we considered the power and memory limitation of smartphone, and for each algorithm we provided two versions: the light and the heavy version. This is a leap forward from previous approaches, which relied on custom-designed sensing and computing platforms. We highlighted how machine learning on smartphones comes with severe costs that need to be mitigated in order to make smartphone capable of real-time data-intensive processing. Some of the costs can be managed with an adapting re-design of the off-the-shelf processing pipeline with additional real-time hyper-parameter control parameters to control the precision and computation cost of the pipeline respect to available resource smartphone in terms of battery duration. We showed that some of the limitations imposed by a mobile sensing application can be overcome by having a multi-tier framework allowing us to split the computation pipeline between the smartphone and two other tiers namely extension-board and cloud, by identifying the bottlenecks in the computation graph. We showed that computation blocks can be can be adopted at execution time leading to further improvement in the resource consumption while maintaining the algorithm accuracy and yet shortening the computation time. We reported on our experience deploying ORBIT at scale with a few case studies as well as multiple deployments on active volcanos in Ecuador and Chile. We extended the scope of our work from platforms to application and presented SPOT. SPOT aims to address some of the challenges discovered in mobile-based smart-home systems. These challenges prevent us from achieving the promises of smart-homes due to heterogeneity in different aspects of smart devices and the underlining systems. This owes to lack of dominating standards in smart-home technologies, leading to the fragmented digital homes rather than truly smart homes. We face the following major heterogeneities in building smart-homes:: (i) Diverse appliance control apps (ii) Communication interface, (iii) Programming abstraction. SPOT is an enabling technology for smart-homes system that allows the integration of hetrogenious smart-device seemless by proposing a novel dynamic draver loading schema. SPOT introduces two driver models namely XML-based and library-based allowing the integration and manipulation of smart devices easy for both programmers and users. SPOT makes the heterogeneous characteristics of smart appliances transparent, and by that minimizes the burden of home automation application developers and the efforts of users who would otherwise have to deal with appliance-specific apps and control interfaces. SPOT is evaluated through several benchmarks and three case studies: cross-device programming, central usage analytics and residential energy management via demand response commands. Our evaluation demonstrates the generality of SPOT’s design and its driver model. After discussing two aspects of this dissertation namely the framework and the application, we finally presented the algorithmic aspect of the dissertation by introducing two systems in smartphone-based deep learning area: Deep-Crowd-Label and Deep-Partition. Deep neural models are both computationally and memory intensive, making them difficult to deploy on mobile applications with limited hardware resources. On the other hand, they are the most advanced machine learning algorithms suitable for real-time sensing applications used in the wild. Deep- Partition is an optimization based partitioning meta-algorithm featuring a tiered architecture for smartphone and the back-end cloud, which helps to deploy and execute deep neural models more efficiently. Deep-Partition provides a profile-based model partitioning allowing it to intelligently execute the Deep Learning algorithms among the tiers to minimize the smartphone power consumption by minimizing the deep models feed-forward latency. Extensive microbenchmark evaluation and three case studies on representative deep neural models validate the performance gain by Deep-Partition. In addition, we presented Deep-Crowd-Label, a novel algorithm designed for distributed collaborative smartphone systems for crowd-sourcing applications. Deep-Crowd-Label is prototyped for semantically labeling user’s location. Deep-Crowd-Label is a crowd-assisted algorithm that uses crowd-sourcing in both training and inference time. It builds deep convolutional neural models using crowd-sensed images to detect the context (label) of indoor locations. It features domain adaptation and model extension via transfer learning to efficiently build deep models for image labeling. By fully exploiting the pre-trained models and available datasets, Deep-Crowd- Label builds ensemble of models to increase the robustness and improve the accuracy of prediction. Moreover, Deep-Crowd-Label aggregates several individual predictions of images obtained from the same location to infer the contextual label of a location. The prototyped system and the preliminary experiments on 26 different in-door locations show the high accuracy of the model and demonstrates the generality and robustness of the underlying approach. The work presented in this dissertation covers three major facets of data-driven and compute-intensive smartphone-based systems, platforms, applications and algorithms; and helps to spurs a new area of research on smartphone sensing and opens up new directions in mobile computing research.
Show less