Individual Subnetworks Research Articles

The improved backward compatible physics-informed neural networks for reducing error accumulation and applications in data-driven higher-order rogue waves.

Due to the dynamic characteristics of instantaneity and steepness, employing domain decomposition techniques for simulating rogue wave solutions is highly appropriate. Wherein, the backward compatible physics-informed neural network (bc-PINN) is a temporally sequential scheme to solve PDEs over successive time segments while satisfying all previously obtained solutions. In this work, we propose improvements to the original bc-PINN algorithm in two aspects based on the characteristics of error propagation. One is to modify the loss term for ensuring backward compatibility by selecting the earliest learned solution for each sub-domain as pseudo-reference solution. The other is to adopt the concatenation of solutions obtained from individual subnetworks as the final form of the predicted solution. The improved backward compatible PINN (Ibc-PINN) is applied to study data-driven higher-order rogue waves for the nonlinear Schrödinger (NLS) equation and the AB system to demonstrate the effectiveness and advantages. Transfer learning and initial condition guided learning (ICGL) techniques are also utilized to accelerate the training. Moreover, the error analysis is conducted on each sub-domain, and it turns out that the slowdown of Ibc-PINN in error accumulation speed can yield greater advantages in accuracy. In short, numerical results fully indicate that Ibc-PINN significantly outperforms bc-PINN in terms of accuracy and stability without sacrificing efficiency.

Phase Analysis of N-Port Electrical Networks under Interconnections

We study a class of matrix operations induced by n-port electrical network interconnections. We first introduce the frequency-dependent phases of n-port linear time-invariant electrical networks in light of the recently defined phases of complex matrices. Such phase characterization can be used to, for example, qualify the well-known notion of passivity for networks in a way. Next, the intrinsic phase properties of networks under interconnections are investigated. A scalable phase-preserving criterion is presented. The criterion involves only the phase properties of individual sub-networks, under the matrix operations featured by interconnections, whereby the phase range of the newly integrated network can be verified effectively, and more importantly maintains the scalability of analysis.

The Connections Between Founders’ Social Network and Human Capital in Technology-Based New Ventures

Abstract This research examines the relationship between a founding team’s social network and the acquisition of its critical human capital. First, explicitly dealing with technology-based firms, we disaggregate the founder(s) social network into four sub-networks: academic, industry, finance, and personal (family and community). Then, we detail the relationship between these individual sub-networks and the acquisition of both technical and business skilled human capital. Our results confirm that individual sub-networks have a differential effect on acquiring both technical and business human capital.

Residual refinement for interactive skin lesion segmentation

BackgroundImage segmentation is a difficult and classic problem. It has a wide range of applications, one of which is skin lesion segmentation. Numerous researchers have made great efforts to tackle the problem, yet there is still no universal method in various application domains.ResultsWe propose a novel approach that combines a deep convolutional neural network with a grabcut-like user interaction to tackle the interactive skin lesion segmentation problem. Slightly deviating from grabcut user interaction, our method uses boxes and clicks. In addition, contrary to existing interactive segmentation algorithms that combine the initial segmentation task with the following refinement task, we explicitly separate these tasks by designing individual sub-networks. One network is SBox-Net, and the other is Click-Net. SBox-Net is a full-fledged segmentation network that is built upon a pre-trained, state-of-the-art segmentation model, while Click-Net is a simple yet powerful network that combines feature maps extracted from SBox-Net and user clicks to residually refine the mistakes made by SBox-Net. Extensive experiments on two public datasets, PH2 and ISIC, confirm the effectiveness of our approach.ConclusionsWe present an interactive two-stage pipeline method for skin lesion segmentation, which was demonstrated to be effective in comprehensive experiments.

MKEL: Multiple Kernel Ensemble Learning via Unified Ensemble Loss for Image Classification

In this article, a novel ensemble model, called Multiple Kernel Ensemble Learning (MKEL), is developed by introducing a unified ensemble loss. Different from the previous multiple kernel learning (MKL) methods, which attempt to seek a linear combination of basis kernels as a unified kernel, our MKEL model aims to find multiple solutions in corresponding Reproducing Kernel Hilbert Spaces (RKHSs) simultaneously. To achieve this goal, multiple individual kernel losses are integrated into a unified ensemble loss. Therefore, each model can co-optimize to learn its optimal parameters by minimizing a unified ensemble loss in multiple RKHSs. Furthermore, we apply our proposed ensemble loss into the deep network paradigm and take the sub-network as a kernel mapping from the original input space into a feature space, named Deep-MKEL (D-MKEL). Our D-MKEL model can utilize the diversified deep individual sub-networks into a whole unified network to improve the classification performance. With this unified loss design, our D-MKEL model can make our network much wider than other traditional deep kernel networks and more parameters are learned and optimized. Experimental results on several mediate UCI classification and computer vision datasets demonstrate that our MKEL model can achieve the best classification performance among comparative MKL methods, such as Simple MKL, GMKL, Spicy MKL, and Matrix-Regularized MKL. On the contrary, experimental results on large-scale CIFAR-10 and SVHN datasets concretely show the advantages and potentialities of the proposed D-MKEL approach compared to state-of-the-art deep kernel methods.

Adversarial Kendall’s Model Towards Containment of Distributed Cyber-Threats

This work examines propagation of cyber-threats over networks under an adversarial formulation. Exploiting Kendall's birth-death-immigration model, we propose an analytical framework to describe the stochastic dynamics of cyber-threat propagation in a collection of heterogeneous sub-networks characterized by different attributes. We propose two formalisations of the problem as zero-sum games involving two adversaries: an attacker, who launches cyber-threats across the distinct sub-networks; and a defender, who tries to mitigate the threats by delivering suitable countermeasures. According to the first formalisation, the interplay between the defender and the attacker is modelled as a Stackelberg leader-follower game, while the second formalisation considers a strategic game wherein the two contenders play simultaneously without knowing the choice of the other player. We derive the equilibrium strategies for both versions of the game, and discuss a number of insightful interplays and ramifications of the different equilibrium points for the problem at hand. The equilibrium strategies depend on three fundamental attributes: i) the available resource budget of the attacker and the defender; ii) the capacity of the legitimate nodes to (unintentionally) forward the threat across the network, after they have been compromised during the propagation of the threat; iii) the intrinsic characteristics of the sub-networks, namely, their immunity to the attacks, their inertia in responding to the countermeasures, and the importance of the individual sub-networks. The relevance of the proposed solution is illustrated through a series of examples and numerical simulations.

Analysis of the air cargo transport network using a complex network theory perspective

In this paper, we present a complex network analysis of the air transport network using the air cargo, instead of the passenger, perspective. To the best of our knowledge, this is the first work where a global cargo network comprising passenger airlines, full-cargo airlines, and integrators’ capacity was studied. We used estimated yearly cargo capacity between airport pairs as input to the model. After assessing network characteristics of the sub-networks representing different carrier types, the full network was obtained as a super-imposition of the individual sub-networks. The resulting network has both small-world and scale-free characteristics. Its topological properties resulted in a higher flow imbalance and concentration with respect to its passenger counterpart, with a smaller characteristic path length and diameter. This result is consistent with the larger catchment area of cargo airports, which heavily rely on road feeder services for the ground leg. Finally, we showed how different attack strategies result in hubs of hub-and-spoke systems or airports behaving as bridges between communities being attacked first. We believe this work to be of relevance both for academics and for practitioners in an era where, due to the soaring of e-commerce and next day delivery, new players are entering the air cargo business and competition is constantly increasing.

Human brain networks: a graph theoretical analysis of cortical connectivity normative database from EEG data in healthy elderly subjects.

Moving from the hypothesis that aging processes modulate brain connectivity networks, 170 healthy elderly volunteers were submitted to EEG recordings in order to define age-related normative limits. Graph theory functions were applied to exact low-resolution electromagnetic tomography on cortical sources in order to evaluate the small-world parameter as a representative model of network architecture. The analyses were carried out in the whole brain-as well as for the left and the right hemispheres separately-and in three specific resting state subnetworks defined as follows: attentional network (AN), frontal network (FN), and default mode network (DMN) in the EEG frequency bands (delta, theta, alpha 1, alpha 2, beta 1, beta 2, gamma). To evaluate the stability of the investigated parameters, a subgroup of 32 subjects underwent three separate EEG recording sessions in identical environmental conditions after a few days interval. Results showed that the whole right/left hemispheric evaluation did not present side differences, but when individual subnetworks were considered, AN and DMN presented in general higher SW in low (delta and/or theta) and high (gamma) frequency bands in the left hemisphere, while for FN, the alpha 1 band was lower in the left with respect to the right hemisphere. It was also evident the test-retest reliability and reproducibility of the present methodology when carried out in clinically stable subjects.Evidences from the present study suggest that graph theory represents a reliable method to address brain connectivity patterns from EEG data and is particularly suitable to study the physiological impact of aging on brain functional connectivity networks.

Architecture and dynamics of overlapped RNA regulatory networks.

A single protein can bind and regulate many mRNAs. Multiple proteins with similar specificities often bind and control overlapping sets of mRNAs. Yet little is known about the architecture or dynamics of overlapped networks. We focused on three proteins with similar structures and related RNA-binding specificities—Puf3p, Puf4p, and Puf5p of S. cerevisiae. Using RNA Tagging, we identified a “super-network” comprised of four subnetworks: Puf3p, Puf4p, and Puf5p subnetworks, and one controlled by both Puf4p and Puf5p. The architecture of individual subnetworks, and thus the super-network, is determined by competition among particular PUF proteins to bind mRNAs, their affinities for binding elements, and the abundances of the proteins. The super-network responds dramatically: The remaining network can either expand or contract. These strikingly opposite outcomes are determined by an interplay between the relative abundance of the RNAs and proteins, and their affinities for one another. The diverse interplay between overlapping RNA–protein networks provides versatile opportunities for regulation and evolution.

Stability switches and bifurcation in a system of four coupled neural networks with multiple time delays

This paper reveals the dynamical behaviors of a coupled network consisting of four neural sub-networks and multiple two-way couplings of neurons between the individual sub-networks. Different types of time delays are introduced into the internal connections within the individual sub-networks and the couplings between the sub-networks. Stability switches of the network equilibrium and Hopf bifurcation are analyzed by discussing the associated characteristic equation. The conditions for the existence of different patterns of oscillations are discussed. By using the Lyapunov’s second method, the global stability of the network equilibrium is studied. Numerical simulations are performed to validate the theoretical results, and rich dynamical behaviors are observed, such as synchronous oscillations, multiple stability switches between the rest state and oscillations, phase-locked oscillations, asynchronous oscillations, and the coexistence of different patterns of bifurcated oscillations.

Brain Signature Predicts Negative Emotion in Individuals.

Emotions are central to our daily lives and are at the root of disorders such as anxiety, depression, drug abuse, and cardiovascular disease. The importance of understanding the emotional brain has motivated hundreds of neuroimaging studies in healthy humans and patients with psychological disorders. But there has been a pressing need to identify neural signatures that predict emotional experiences in individuals in order to translate brain-imaging data into clinical applications. In a study published this week in PLOS Biology, Luke Chang and Tor Wager of the University of Colorado at Boulder and their collaborators used functional magnetic resonance imaging (fMRI) and machine-learning algorithms to identify a sensitive and specific brain signature that predicts the intensity of negative emotion in individuals [1]. Strikingly, this signature encompasses multiple widely distributed neural networks rather than a few individual emotion-related brain regions. According to the authors, the neural signature could be used to diagnose mental illness, predict responses to treatment, and guide the development of new brain-based therapies. In the new study, the researchers used fMRI to measure brain activity in 183 individuals who viewed a sequence of 15 negative photographs and 15 neutral photographs and rated the emotional intensity of each image on a five-point scale. The negative photographs depicted bodily illness and injury, acts of aggression, members of hate groups, transportation accidents, and human waste, whereas the neutral photographs consisted of inanimate objects or nonevocative scenes (Fig 1). Fig 1 Luke Chang, Tor Wager, and colleagues used images similar to this to elicit negative emotions in the study participants while their brains were being scanned. The researchers then used machine-learning algorithms to find global patterns of brain activity that best predicted participants’ ratings. This neural signature, called the picture-induced negative emotion signature (PINES), accurately predicted ratings of negative emotional experience in 94% of the participants. In more than 90% of the subjects, the PINES accurately distinguished between aversive pictures whose ratings were separated by two or more points on the five-point scale. As expected, aversive pictures produced activity in brain regions typically associated with negative emotion, including the amygdala, anterior insula, and anterior cingulate cortex. However, activity in each of these regions was not strong enough to predict emotion ratings. Moreover, the PINES more accurately predicted emotion ratings than did activity in seven other previously reported neural networks associated with diverse functions such as rest, vision, attention, touch, and movement. The researchers next used algorithms to separate the PINES brain regions into discrete subnetworks based on similar patterns of activity from trial to trial. They identified nine subnetworks involved in diverse functions such as vision, arousal, memory, social cognition, and action. Several of these subnetworks individually performed well at predicting emotion ratings, but all of them were less accurate than the PINES. Using a virtual lesion analysis, the researchers found that removing individual subnetworks from the PINES resulted in only negligible decreases in performance. The findings demonstrate that no specific subsystem is either necessary or sufficient for predicting negative emotion. Finally, the researchers determined whether the PINES is specific to negative emotion and is not driven simply by general arousal. To do so, they compared the PINES to their previously reported neurologic pain signature (NPS)—a brain map that predicts the intensity of thermal pain. Similar to the PINES, the NPS was generated by a combination of fMRI and machine-learning algorithms. Comparison of the two signatures revealed that the experiences of negative emotion and pain are predicted by distinct patterns of neural activity. The PINES accurately predicted negative emotion but showed no response to increasing pain intensity. Conversely, the NPS responded robustly to increasing pain but showed no response to increasing negative emotion. Taken together, the results suggest that the neural signature predicting negative emotion spans multiple brain systems and cannot be boiled down to a single network or a few brain regions that are traditionally associated with negative emotion. Therefore, the study calls into question theories that have treated emotion as an isolated experience that is localizable to a specific neural system. By providing an alternative, more nuanced model for the brain representation of emotion, the findings have broad implications for future attempts to identify biomarkers for mental health disorders.

A Low Complexity Data Driven Model of Environmental Discharge Dynamics for Wireless Sensor Network Applications

Poor water quality is a global concern, with agricultural practices the major contributor to reduced water quality with emissions of nutrient fluxes in to water systems. Using a collaborative framework to support catchment-scale water quality monitoring, control and management (WQMCM), individual sub-networks can learn and predict the impact of catchment events on their locality[1], allowing dynamic decision making for local irrigation strategies. Since resource constraints on network nodes (e.g. battery life, computing power etc) require a simplified predictive model, low-dimensional model parameters are derived from the existing National Resource Conservation Method (NRCS). An M5 decision tree algorithm is then used to develop predictive models for total discharge volume (Q), response start and duration (t1 & td). Evaluation of these models demonstrates high accuracy (84-94%) even for a small training set of under 100 samples for Q and td. However, for t1, 300 samples are required to give adequate performance.

Exploration of potential roles of a new LOXL2 splicing variant using network knowledge in esophageal squamous cell carcinoma.

LOXL2 (lysyl oxidase-like 2), an enzyme that catalyzes oxidative deamination of lysine residue, is upregulated in esophageal squamous cell carcinoma (ESCC). A LOXL2 splice variant LOXL2-e13 and its wild type were overexpressed in ESCC cells followed by microarray analyses. In this study, we explored the potential role and molecular mechanism of LOXL2-e13 based on known protein-protein interactions (PPIs), following microarray analysis of KYSE150 ESCC cells overexpressing a LOXL2 splice variant, denoted by LOXL2-e13, or its wild-type counterpart. The differentially expressed genes (DEGs) of LOXL2-WT and LOXL2-e13 were applied to generate individual PPI subnetworks in which hundreds of DEGs interacted with thousands of other proteins. These two DEG groups were annotated by Functional Annotation Chart analysis in the DAVID bioinformatics database and compared. These results found many specific annotations indicating the potential specific role or mechanism for LOXL2-e13. The DEGs of LOXL2-e13, comparing to its wild type, were prioritized by the Random Walk with Restart algorithm. Several tumor-related genes such as ERO1L, ITGA3, and MAPK8 were found closest to LOXL2-e13. These results provide helpful information for subsequent experimental identification of the specific biological roles and molecular mechanisms of LOXL2-e13. Our study also provides a work flow to identify potential roles of splice variants with large scale data.

Causal inference of gene regulation with subnetwork assembly from genetical genomics data

Deciphering the causal networks of gene interactions is critical for identifying disease pathways and disease-causing genes. We introduce a method to reconstruct causal networks based on exploring phenotype-specific modules in the human interactome and including the expression quantitative trait loci (eQTLs) that underlie the joint expression variation of each module. Closely associated eQTLs help anchor the orientation of the network. To overcome the inherent computational complexity of causal network reconstruction, we first deduce the local causality of individual subnetworks using the selected eQTLs and module transcripts. These subnetworks are then integrated to infer a global causal network using a random-field ranking method, which was motivated by animal sociology. We demonstrate how effectively the inferred causality restores the regulatory structure of the networks that mediate lymph node metastasis in oral cancer. Network rewiring clearly characterizes the dynamic regulatory systems of distinct disease states. This study is the first to associate an RXRB-causal network with increased risks of nodal metastasis, tumor relapse, distant metastases and poor survival for oral cancer. Thus, identifying crucial upstream drivers of a signal cascade can facilitate the discovery of potential biomarkers and effective therapeutic targets.

Stability switches, bifurcation, and multi-stability of coupled networks with time delays

This paper reveals the dynamical behaviors of a network consisting of a pair of ring sub-networks and two-way couplings of neurons between the individual sub-networks. Time delays are introduced into the internal connections within the individual sub-networks and the couplings between the individual sub-networks. The stability and instability of the trivial equilibrium of the network are analyzed and the conditions for the existence of Hopf bifurcation are obtained by discussing the associated characteristic equation. The criterion for the global stability of the trivial equilibrium of the network is given by constructing a suitable Lyapunov functional. Numerical simulations are performed to validate the theoretical results and rich dynamical behaviors are observed, such as multiple stability switches of the network equilibrium, synchronous/asynchronous oscillations, and multi-stability.

An Arabidopsis gene network based on the graphical Gaussian model

We describe a gene network for the Arabidopsis thaliana transcriptome based on a modified graphical Gaussian model (GGM). Through partial correlation (pcor), GGM infers coregulation patterns between gene pairs conditional on the behavior of other genes. Regularized GGM calculated pcor between gene pairs among approximately 2000 input genes at a time. Regularized GGM coupled with iterative random samplings of genes was expanded into a network that covered the Arabidopsis genome (22,266 genes). This resulted in a network of 18,625 interactions (edges) among 6760 genes (nodes) with high confidence and connections representing approximately 0.01% of all possible edges. When queried for selected genes, locally coherent subnetworks mainly related to metabolic functions, and stress responses emerged. Examples of networks for biochemical pathways, cell wall metabolism, and cold responses are presented. GGM displayed known coregulation pathways as subnetworks and added novel components to known edges. Finally, the network reconciled individual subnetworks in a topology joined at the whole-genome level and provided a general framework that can instruct future studies on plant metabolism and stress responses. The network model is included.

Inner and minimum constraint adjustment of marine gravity data

A programme for marine gravity data validation and adjustment is presented—VALDAMA. The aim of the programme is to provide a complete user-friendly system for marine gravity data validation and adjustment that enables the user to define all intervening parameters. The programme is written in standard C and has the possibility of solving for three different types of adjustment: inner constraint, constraint and minimum constraint and allows the user to define the stochastic model. It also has the ability to detect and adjust individual sub-networks, which may occur in regional applications, where track network connectivity fails. VALDAMA uses a parameter file in which the validation parameters, adjustment method and stochastic model are defined. The programme calculates the main statistical parameters and presents the results in spreadsheet format compatible with most common computer office tools. A test case is presented for marine gravity data in the North-East Atlantic Ocean. The complete validation and adjustment of more than 190,000 gravity observations are presented and from an analysis of several adjustment possibilities, it was concluded that inner constraint adjustment may give unpredictable results for global gravity data and that minimum constraint would be the preferred solution.

Product modeling to support case-based construction planning and scheduling

Many human schedulers create schedules by reusing past similar schedules. The retrieval and reuse of similar schedules are subjective and experience-based. This paper explores different notions of similarity required when performing different scheduling tasks. It describes the CasePlan system that helps schedulers retrieve and reuse parts of existing schedules based on a generic product model, and apply case-based reasoning to generate new schedules. The validation experiment demonstrated CasePlan's accuracy in determining individual subnetworks and activity durations, but weak performance in determining interlinks between subnetworks, which highly depends upon the availability of pertinent cases and the level of detail of project information.

Modelling end-to-end protocols over interconnected heterogeneous networks

The performance aspects of end-to-end communication in interconnected heterogeneous networks are investigated. The individual networks to be connected (e.g. LANs, high-speed LANs, X.25 networks, WANs including satellite links, ISDN, B-ISDN) differ significantly in their quality of service. On the other hand, requirements imposed by applications also differ significantly. Thus there is no single protocol stack which is optimum for all cases, and decisions have to be made on a special case-by-case basis. Motivated by the planned European space mission, for which a proper protocol stack assuring reliable end-to-end communication between application processes residing in space labs and ground-based application processes has to be chosen, a novel, top down modelling approach is developed. Rather than combining models of individual subnetworks, we represent any type of subnetwork by a specific set of networl quality of service characteristics, and model the transport protocol so as to express the obtainable transport quality of service. Using such an approach it is possible to decide what kind of technological constraints have to be imposed on the parameters of a subnetwork A, which is to be connected to the given subnetworks B and C so as to assure the desired end-to-end quality of service.

Multilayer Control of Sewer Networks

A multilayer control system for sewer networks combining optimal flow diversion for overflow minimization with moderate implementation cost is presented. The multilayer structuic consist s of an opt imizat ion layer, a direct control layer and an adaption layer. Control problems connected with these control layers are described. The discrete maximum principle and conjugate gradient techniques are applied for the solution of a simplified optimization problem for the overall network. Optimization results are used as reference trajectories for classical closed-loop control of individual subnetworks. Predicted inflow trajectories needed for the optimization procedure are provided by the adaptation layer.