ASU Scholarship Showcase

Two classes of scaling behaviours, namely the super-linear scaling of links or activities, and the sub-linear scaling of area, diversity, or time elapsed with respect to size have been found to prevail in the growth of complex networked systems. Despite some pioneering modelling approaches proposed for specific systems, whether there exists some general mechanisms that account for the origins of such scaling behaviours in different contexts, especially in socioeconomic systems, remains an open question. We address this problem by introducing a geometric network model without free parameter, finding that both super-linear and sub-linear scaling behaviours can be simultaneously reproduced and that the scaling exponents are exclusively determined by the dimension of the Euclidean space in which the network is embedded. We implement some realistic extensions to the basic model to offer more accurate predictions for cities of various scaling behaviours and the Zipf distribution reported in the literature and observed in our empirical studies. All of the empirical results can be precisely recovered by our model with analytical predictions of all major properties. By virtue of these general findings concerning scaling behaviour, our models with simple mechanisms gain new insights into the evolution and development of complex networked systems.

The number and variety of connectivity estimation methods is likely to continue to grow over the coming decade. Comparisons between methods are necessary to prune this growth to only the most accurate and robust methods. However, the nature of connectivity is elusive with different methods potentially attempting to identify different aspects of connectivity. Commonalities of connectivity definitions across methods upon which base direct comparisons can be difficult to derive. Here, we explicitly define “effective connectivity” using a common set of observation and state equations that are appropriate for three connectivity methods: dynamic causal modeling (DCM), multivariate autoregressive modeling (MAR), and switching linear dynamic systems for fMRI (sLDSf). In addition while deriving this set, we show how many other popular functional and effective connectivity methods are actually simplifications of these equations. We discuss implications of these connections for the practice of using one method to simulate data for another method. After mathematically connecting the three effective connectivity methods, simulated fMRI data with varying numbers of regions and task conditions is generated from the common equation. This simulated data explicitly contains the type of the connectivity that the three models were intended to identify. Each method is applied to the simulated data sets and the accuracy of parameter identification is analyzed. All methods perform above chance levels at identifying correct connectivity parameters. The sLDSf method was superior in parameter estimation accuracy to both DCM and MAR for all types of comparisons.

The estimation of energy demand (by power plants) has traditionally relied on historical energy use data for the region(s) that a plant produces for. Regression analysis, artificial neural network and Bayesian theory are the most common approaches for analysing these data. Such data and techniques do not generate reliable results. Consequently, excess energy has to be generated to prevent blackout; causes for energy surge are not easily determined; and potential energy use reduction from energy efficiency solutions is usually not translated into actual energy use reduction. The paper highlights the weaknesses of traditional techniques, and lays out a framework to improve the prediction of energy demand by combining energy use models of equipment, physical systems and buildings, with the proposed data mining algorithms for reverse engineering. The research team first analyses data samples from large complex energy data, and then, presents a set of computationally efficient data mining algorithms for reverse engineering. In order to develop a structural system model for reverse engineering, two focus groups are developed that has direct relation with cause and effect variables. The research findings of this paper includes testing out different sets of reverse engineering algorithms, understand their output patterns and modify algorithms to elevate accuracy of the outputs.

Small and medium office buildings consume a significant parcel of the U.S. building stock energy consumption. Still, owners lack resources and experience to conduct detailed energy audits and retrofit analysis. We present an eight-steps framework for an energy retrofit assessment in small and medium office buildings. Through a bottom-up approach and a web-based retrofit toolkit tested on a case study in Arizona, this methodology was able to save about 50% of the total energy consumed by the case study building, depending on the adopted measures and invested capital. While the case study presented is a deep energy retrofit, the proposed framework is effective in guiding the decision-making process that precedes any energy retrofit, deep or light.

Commercial buildings’ consumption is driven by multiple factors that include occupancy, system and equipment efficiency, thermal heat transfer, equipment plug loads, maintenance and operational procedures, and outdoor and indoor temperatures. A modern building energy system can be viewed as a complex dynamical system that is interconnected and influenced by external and internal factors. Modern large scale sensor measures some physical signals to monitor real-time system behaviors. Such data has the potentials to detect anomalies, identify consumption patterns, and analyze peak loads. The paper proposes a novel method to detect hidden anomalies in commercial building energy consumption system. The framework is based on Hilbert-Huang transform and instantaneous frequency analysis. The objectives are to develop an automated data pre-processing system that can detect anomalies and provide solutions with real-time consumption database using Ensemble Empirical Mode Decomposition (EEMD) method. The finding of this paper will also include the comparisons of Empirical mode decomposition and Ensemble empirical mode decomposition of three important type of institutional buildings.

There are many data mining and machine learning techniques to manage large sets of complex energy supply and demand data for building, organization and city. As the amount of data continues to grow, new data analysis methods are needed to address the increasing complexity. Using data from the energy loss between the supply (energy production sources) and demand (buildings and cities consumption), this paper proposes a Semi-Supervised Energy Model (SSEM) to analyse different loss factors for a building cluster. This is done by deep machine learning by training machines to semi-supervise the learning, understanding and manage the process of energy losses. Semi-Supervised Energy Model (SSEM) aims at understanding the demand-supply characteristics of a building cluster and utilizes the confident unlabelled data (loss factors) using deep machine learning techniques. The research findings involves sample data from one of the university campuses and presents the output, which provides an estimate of losses that can be reduced. The paper also provides a list of loss factors that contributes to the total losses and suggests a threshold value for each loss factor, which is determined through real time experiments. The conclusion of this paper provides a proposed energy model that can provide accurate numbers on energy demand, which in turn helps the suppliers to adopt such a model to optimize their supply strategies.

Background: Carriers of the APOE ε4 allele are at increased risk of developing Alzheimer’s disease (AD), and have been shown to have reduced cerebral metabolic rate of glucose (CMRgl) in the same brain areas frequently affected in AD. These individuals also exhibit reduced plasma levels of apolipoprotein E (apoE) attributed to a specific decrease in the apoE4 isoform as determined by quantification of individual apoE isoforms in APOE ε4 heterozygotes. Whether low plasma apoE levels are associated with structural and functional brain measurements and cognitive performance remains to be investigated.

Methods: Using quantitative mass spectrometry we quantified the plasma levels of total apoE and the individual apoE3 and apoE4 isoforms in 128 cognitively normal APOE ε3/ε4 individuals included in the Arizona APOE cohort. All included individuals had undergone extensive neuropsychological testing and 25 had in addition undergone FDG-PET and MRI to determine CMRgl and regional gray matter volume (GMV).

Results: Our results demonstrated higher apoE4 levels in females versus males and an age-dependent increase in the apoE3 isoform levels in females only. Importantly, a higher relative ratio of apoE4 over apoE3 was associated with GMV loss in the right posterior cingulate and with reduced CMRgl bilaterally in the anterior cingulate and in the right hippocampal area. Additional exploratory analysis revealed several negative associations between total plasma apoE, individual apoE isoform levels, GMV and CMRgl predominantly in the frontal, occipital and temporal areas. Finally, our results indicated only weak associations between apoE plasma levels and cognitive performance which further appear to be affected by sex.

Conclusions: Our study proposes a sex-dependent and age-dependent variation in plasma apoE isoform levels and concludes that peripheral apoE levels are associated with GMV, CMRgl and possibly cognitive performance in cognitively healthy individuals with a genetic predisposition to AD.

Background: We introduced a hypometabolic convergence index (HCI) to characterize in a single measurement the extent to which a person’s fluorodeoxyglucose positron emission tomogram (FDG PET) corresponds to that in Alzheimer’s disease (AD). Apolipoprotein E ε4 (APOE ε4) gene dose is associated with three levels of risk for late-onset AD. We explored the association between gene dose and HCI in cognitively normal ε4 homozygotes, heterozygotes, and non-carriers.

Methods: An algorithm was used to characterize and compare AD-related HCIs in cognitively normal individuals, including 36 ε4 homozygotes, 46 heterozygotes, and 78 non-carriers.

Results: These three groups differed significantly in their HCIs (ANOVA, p = 0.004), and there was a significant association between HCIs and gene dose (linear trend, p = 0.001).

Conclusions: The HCI is associated with three levels of genetic risk for late-onset AD. This supports the possibility of using a single FDG PET measurement to help in the preclinical detection and tracking of AD.

Given a complex geospatial network with nodes distributed in a two-dimensional region of physical space, can the locations of the nodes be determined and their connection patterns be uncovered based solely on data? We consider the realistic situation where time series/signals can be collected from a single location. A key challenge is that the signals collected are necessarily time delayed, due to the varying physical distances from the nodes to the data collection centre. To meet this challenge, we develop a compressive-sensing-based approach enabling reconstruction of the full topology of the underlying geospatial network and more importantly, accurate estimate of the time delays. A standard triangularization algorithm can then be employed to find the physical locations of the nodes in the network. We further demonstrate successful detection of a hidden node (or a hidden source or threat), from which no signal can be obtained, through accurate detection of all its neighbouring nodes. As a geospatial network has the feature that a node tends to connect with geophysically nearby nodes, the localized region that contains the hidden node can be identified.

Recent works revealed that the energy required to control a complex network depends on the number of driving signals and the energy distribution follows an algebraic scaling law. If one implements control using a small number of drivers, e.g. as determined by the structural controllability theory, there is a high probability that the energy will diverge. We develop a physical theory to explain the scaling behaviour through identification of the fundamental structural elements, the longest control chains (LCCs), that dominate the control energy. Based on the LCCs, we articulate a strategy to drastically reduce the control energy (e.g. in a large number of real-world networks). Owing to their structural nature, the LCCs may shed light on energy issues associated with control of nonlinear dynamical networks.