Master of Science, The Ohio State University, 2017, Mechanical Engineering

With sporadic advancement in computer technology, transportation is moving towards autonomy. With rapid increase in production of highly automated vehicles (AVs), validation and safety of AVs is gaining high importance. The estimation of safety for AVs is a challenging problem as the AVs mimic human drivers and it requires an estimate of AVs response at all critical scenarios. AV response in each scenario, if known, can be used for estimating its safety. In this work, methods for estimating vehicle response are proposed by using various models based on both physics-based modeling as well as Machine Learning algorithms. Various Machine Learning algorithms were explored for classifying and predicting driver's intention, such as Extremely Randomized Trees and Gaussian Mixture Model based Hidden Markov Model. Also, physics-based modeling is done for longitudinal car-following conditions using three models namely: Spring-Damper model, Time-to-Collision model and Gazis-Herman-Rothery model. The Machine Learning models were fitted using Naturalistic Driving Study dataset (NDS) collected as a part of Strategic Highway Research Program-2 (SHRP2). The vehicular data comprising of various vehicular parameters is processed and analyzed for preparing driver's behavior model, which gives an estimate of vehicle's longitudinal and lateral acceleration at that given instance. Physics-based models were limited to longitudinal acceleration prediction as lateral acceleration prediction in dynamic traffic conditions is a highly complex problem for modeling. The physics-based models were fitted using both SHRP2 as well as the test track data of AVs collected from Transportation Research Center Inc. Then, the fitted Machine Learning and physics-based models were validated against validation data. The parameters obtained from physics-based models were used for obtaining driving characteristics, which were used to compare tested AVs among themselves as well as human drivers.

Committee: Gary Heydinger (Advisor); Dennis Guenther (Committee Chair) Subjects: Engineering; Mechanical Engineering; Transportation

Doctor of Philosophy, The Ohio State University, 2019, Microbiology

The Arctic system has been undergoing a rampant change during the Anthropocene. This anthropogenic change has allowed for additional physical and biological positive feedback processes that in turn accelerate warming in the arctic and subarctic systems. Microbial community/functional dynamics are both (i) dramatically impacted by these rapid changes and (ii) key players in the biological positive feedback process that accelerates the change. Recent technological, analytical, and computational advances have allowed us to ask systems-level questions that encompass microbial and viral community dynamics (along with their potential functional dynamics) and high-resolution environmental measurements. This research took a systems-level approach to look for the first time at (i) the characteristics of Arctic marine viruses in a global context, and (ii) microbial community gene expression in a rapidly changing permafrost thaw gradient. Additionally, novel viral sequences recovered from the marine and terrestrial ecosystems studied here were used to build new resources and tools that accelerate viral discovery in nature. First, studying marine viral macro- and microdiversity from the Arctic Ocean to the Southern Ocean, enabled by the Tara Oceans Expedition, revealed the Arctic Ocean as a hotspot of viral diversity, with ~42% of the recovered viral populations originating from the Arctic Ocean viromes. In total 195,728 viral populations >10 kb were recovered from the global ocean to constitute the Global Ocean Viromes 2.0 (GOV2.0) dataset. Viral communities assorted into five distinct global ecological zones and the arctic viral communities formed their own distinct ecological zone. Additionally, this work revealed unexpected patterns and ecological drivers of viral diversity (at the community, inter-, and intrapopulation levels), within the Arctic Ocean, across latitudes, and across the depth of the global ocean. Second, genome-resolved metaproteomic study of microbial gene (open full item for complete abstract)

Committee: Matthew Sullivan (Advisor); Virginia Rich (Advisor); Kelly Wrighton (Committee Member); Alvaro Montenegro (Committee Member) Subjects: Biogeochemistry; Bioinformatics; Biological Oceanography; Biology; Climate Change; Ecology; Environmental Science; Geobiology; Microbiology; Oceanography; Soil Sciences; Statistics; Virology

Doctor of Philosophy (PhD), Wright State University, 2014, Computer Science and Engineering PhD

Study of the human brain is an important and very active area of research. Unraveling the way the human brain works would allow us to better understand, predict and prevent brain related diseases that affect a significant part of the population. Studying the brain response to certain input stimuli can help us determine the involved brain areas and understand the mechanisms that characterize behavioral and psychological traits. In this research work two methods used for the monitoring of brain activities, Electroencephalography (EEG) and functional Magnetic Resonance (fMRI) have been studied for their fusion, in an attempt to bridge together the advantages of each one. In particular, this work has focused in the analysis of a specific type of EEG and fMRI recordings that are related to certain events and capture the brain response under specific experimental conditions. Using spatial features of the EEG we can describe the temporal evolution of the electrical field recorded in the scalp of the head. This work introduces the use of Hidden Markov Models (HMM) for modeling the EEG dynamics. This novel approach is applied for the discrimination of normal and progressive Mild Cognitive Impairment patients with significant results. EEG alone is not able to provide the spatial localization needed to uncover and understand the neural mechanisms and processes of the human brain. Functional Magnetic Resonance imaging (fMRI) provides the means of localizing functional activity, without though, providing the timing details of these activations. Although, at first glance it is apparent that the strengths of these two modalities, EEG and fMRI, complement each other, the fusion of information provided from each one is a challenging task. A novel methodology for fusing EEG spatiotemporal features and fMRI features, based on Canonical Partial Least Squares (CPLS) is presented in this work. A HMM modeling approach is used in order to derive a novel feature-based representation of (open full item for complete abstract)

Committee: Nikolaos Bourbakis Ph.D. (Advisor); Soon Chung Ph.D. (Committee Member); Yong Pei Ph.D. (Committee Member); Larry Lawhorne M.D. (Committee Member); Michalis Zervakis Ph.D. (Committee Member) Subjects: Biomedical Research; Computer Engineering

PhD, University of Cincinnati, 2009, Engineering : Computer Science and Engineering

In a cognitive radio network (CRN), bands of a spectrum are shared by licensed (primary) and unlicensed (secondary) users in that preferential order. It is generally recognized that the spectral occupancy by primary users exhibit dynamical spatial and temporal properties. In the open literature, there exist no accurate time-varying model representing the spectrum occupancy that the wireless researchers could employ for evaluating new algorithms and techniques designed for dynamic spectrum access (DSA). We use statistical characteristics from actual radio frequency measurements, obtain first- and second-order parameters, and define a statistical spectrum occupancy model based on a combination of several different probability density functions (PDFs). One of the fundamental issues in analyzing spectrum occupancy is to characterize it in terms of probabilities and study probabilistic distributions over the spectrum. To reduce computational complexity of the exact distribution of total number of free bands, we resort to efficient approximation techniques. Furthermore, we characterize free bands into five different types based on the occupancy of its adjacent bands. The probability distribution of total number of each type of bands is therefore determined. Two corresponding algorithms are effectively developed to compute the distributions, and our extensive simulation results show the effectiveness of the proposed analytical model. Design of an efficient spectrum sensing scheme is a challenging task, especially when false alarms and misdetections are present. The status of the band is to be monitored over a number of consecutive time periods, with each time period being of a specific time interval. The status of the sub-band at any time point is either free or busy. We proved that the status of the band over time evolves randomly, following a Markov chain. The cognitive radio assesses the band, whether or not it is free, and the assessment is prone to errors. The errors (open full item for complete abstract)

Committee: Dharma Agrawal (Advisor); Raj Bhatnagar (Committee Member); Chia-Yung Han (Committee Member); Yiming Hu (Committee Member); Marepalli Rao (Committee Member) Subjects: Computer Science

MS, University of Cincinnati, 2005, Engineering : Computer Science

The field of Bio-Informatics is fast growing with research in various related topics. One such topic is protein sequence classification. This thesis uses this topic as motivation to develop a methodology that uses Hidden Markov Models (HMMs) to classify sequences. Hidden Markov Models are a concept in probability theory widely known for their application in the speech recognition. The three phases of HMMs: training, decoding, and evaluation, are used to classify sequences into clusters that have known similar functional properties. The training phase of HMMs uses a cluster of sequences to learn a model that is most likely to generate the sequences in the training cluster. The decoding and evaluation phases of HMMs use the generated model to calculate the likelihood of an unknown sequence belonging to the same sequence and generating a most-probable path the sequence traverses. The thesis presents background on HMMs along with detailed explanations of the algorithms used to implement all three phases of HMMs. The primary focus of this thesis is on the training phase of HMMs. During the implementation of the training phase we discovered that the phase has a numerical and computational weakness relating to those structures in which some silent states are included as part of the model. The results presented in this thesis test the training algorithms, show their workability and weaknesses, and point towards the silent states related weakness.

Committee: Dr. Raj Bhatnagar (Advisor) Subjects:

Doctor of Philosophy, The Ohio State University, 2006, Statistics

Relapsing remitting multiple sclerosis (RRMS) is a chronic and autoimmune disease where the disease states alternate between the relapse and remission. Magnetic resonance imaging (MRI) is widely used to monitor the pathological progression of this disease. The longitudinal T1-weighted Gadolinium-enhancing MRI lesion count sequences provide information on the onset and sojourn time of the lesion enhancement. We construct biologically interpretable queueing models for the longitudinal data of these lesion counts that describe the natural evolution of the lesions. The infinite-server queue with Poisson arrival process and exponential service (M/M/∞) is proposed for this purpose. The rate of the Poisson arrival process can also be allowed to be governed by a two-state hidden Markov chain. We describe the likelihood function for each model based on appropriate assumptions and fit these models to data from 9 RRMS patients. We obtain the maximum likelihood estimators of the parameters of interest arising from these models and study their asymptotic properties through simulation. We discuss the validation of the assumptions for the proposed models and examine the robustness of these estimators. We suggest the application of the models for characterizing the disease progression and testing treatment effect and discuss implication for planning of RRMS clinical trials.

Committee: Haikady Nagaraja (Advisor); Catherine Calder (Other); Kottil Rammohan (Other); Thomas Santner (Other) Subjects: Statistics

Doctor of Philosophy (PhD), Ohio University, 2006, Electrical Engineering & Computer Science (Engineering and Technology)

Video events contain rich semantic information. Using computational approaches to analyze video events is very important for many applications due to the desire to interpret digital data in a way that is consistent with human knowledge. This thesis investigates object-based video event analysis based on a statistical framework. Within the proposed architecture for object-based video event understanding, object detection is addressed by model-based approaches with the integration of prior color/shape knowledge and recognition feedback. Object classification is investigated as shape-based image retrieval. The relevance feedback is used to adaptively derive basis vectors to capture a user's perceptual preferences. The major focus of this thesis is concerned with statistical modeling of facial event recognition. Two hidden Markov model (HMM) based approaches are presented. The first approach tracks the deformation of facial components in image sequences via active shape models (ASMs) and extracts geometric-based features for facial gestures. The interaction between upper and lower facial components is explicitly modeled via coupled HMMs (CHMMs) by introducing coupled dependencies between hidden variables. The second approach automatically locates face regions in each image frame via eigenanalysis, extracts multi-band appearance features based on Gabor filtering, and models the spatio-temporal stochastic structure of facial image sequences using hierarchical HMMs (HHMMs). The major contributions of this thesis include: 1) a fully automatic person-independent facial expression recognition prototype system; 2) modeling of the spatio-temporal structure of facial image sequences within a hierarchical framework; 3) derived generalized inference and learning algorithms of HHMMs for observation sequences with known multi-scale structures; 4) improved performance of ASMs for facial component tracking by using dynamic programming based search with contextual constraints; 5) expli (open full item for complete abstract)

Committee: David Chelberg (Advisor) Subjects:

Doctor of Philosophy, The Ohio State University, 2013, Electrical and Computer Engineering

Cyber-physical systems (CPS) refer to the co-joining of environmental and computational elements of a system. One CPS application area is in autonomous vehicles. Autonomous (or self-driving) vehicles are likely to be an upcoming revolution in personal and commercial transportation. While there are many outstanding public policy questions, this technology promises to improve our quality of life by providing transportation that is safe and efficient. A likely technology adoption path includes a period in which human driven and autonomous vehicles will need to coexist. In such an environment, referred to as a Mixed Urban Environment, autonomous vehicles may only be able to obtain information from human driven vehicles through on board sensors or vehicle-to-vehicle communication. From this information, an autonomous vehicle will need to determine the likely behavior of the human driven vehicle, a task which is referred to as driver behavior estimation. This task requires a qualitative-quantitative architecture capable of explaining the driver/vehicle coupling being observed. A vehicle's ability to determine other vehicle's likely behavior also has applications to driver safety and collision avoidance systems. In essence, a vehicle must be able to estimate the behavior of another vehicle, and determine its course of action. This thesis proposes an architecture for driver behavior estimation through the unified development of two theoretical concepts, namely: Graphical models, and Hybrid State Systems. Hybrid State Systems (HSS) provide the qualitative relationship between driver/vehicle couplings through a two layer model. Pattern recognition techniques in conjunction with Hidden Markov Models (HMMs), a type of graphical model, provide the quantitative relation between HSS layers. The estimation of current driver state is based on easy-to-measure continuous observations. The proposed system uses machine-learning concepts and requires extensive data collection, which (open full item for complete abstract)

Committee: Ashok Krishnamurthy Dr. (Advisor); Umit Ozguner Dr. (Committee Member); Giorgio Rizzoni Dr. (Committee Member) Subjects: Computer Engineering; Electrical Engineering