Surrogate Networks Research Articles

Voroto: multiscale topology optimization of voronoi structures using surrogate neural networks

Voroto: multiscale topology optimization of voronoi structures using surrogate neural networks

Exploring neural oscillations during speech perception via surrogate gradient spiking neural networks.

Understanding cognitive processes in the brain demands sophisticated models capable of replicating neural dynamics at large scales. We present a physiologically inspired speech recognition architecture, compatible and scalable with deep learning frameworks, and demonstrate that end-to-end gradient descent training leads to the emergence of neural oscillations in the central spiking neural network. Significant cross-frequency couplings, indicative of these oscillations, are measured within and across network layers during speech processing, whereas no such interactions are observed when handling background noise inputs. Furthermore, our findings highlight the crucial inhibitory role of feedback mechanisms, such as spike frequency adaptation and recurrent connections, in regulating and synchronizing neural activity to improve recognition performance. Overall, on top of developing our understanding of synchronization phenomena notably observed in the human auditory pathway, our architecture exhibits dynamic and efficient information processing, with relevance to neuromorphic technology.

Targeted Model Inversion: Distilling style encoded in predictions

Previous model inversion (MI) research has demonstrated the feasibility of reconstructing images representative of specific classes, inadvertently revealing additional feature information. However, there are two remaining challenges for practical black-box MI: (1) minimizing the number of queries to the target model, and (2) reconstructing a high-quality input image tailored to an observed prediction vector. We introduce Targeted Model Inversion (TMI), a practical black-box MI attack. Our approach involves altering the mapping network in StyleGAN, which projects an observed prediction vector into a StyleGAN latent representation. Later, TMI leverages a surrogate model that is also derived from StyleGAN to guide instance-specific MI by optimizing the latent representation. These mapping and surrogate networks work together to conduct high-fidelity MI while significantly decreasing the number of necessary queries. Our experiments demonstrate that TMI outperforms state-of-the-art MI methods, demonstrating a new upper bound on the susceptibility to black-box MI attacks.

Genetic algorithm optimized artificial neural network models of single- and multi-component gas adsorption isotherms for hydrogen purification

Accurate isotherm models contribute to predicting the amount of adsorbate adsorbed in the adsorbent within a certain range of pressure and temperature. Experimental adsorption isotherms of five components (CO2, CO, N2, CH4 and H2) on pelletized zeolite 13X from publication data were correlated with common-used fundamental isotherm models, including Langmuir, Sips, dual-site Langmuir (DSL), and temperature-dependent (TD) isotherm models, including TD Langmuir, TD Sips and TD DSL. Overall, the average relative errors (AREs) of TD models are higher than those of common-used fundamental isotherm models, but TD models have better applicability. In addition, three surrogate artificial neural networks (ANNs) optimized by genetic algorithm (GA) of single- and multi-component gas adsorption isotherm models are proposed. Three GA-ANN models set the adsorption amount as the output and take one, two or three potential inputs, such as adsorption pressure, temperature and adsorbate components. Three GA-ANN models generally have lower AREs and better applicability than fundamental isotherm models and TD isotherm models. The proposed GA-ANN3 model, which has three inputs, makes a new attempt at adsorption isotherms and can fit the multi-component gas adsorption at different temperatures with the change of adsorption pressure. The adsorber dynamics for multi-component adsorption (CO2: CO: N2: H2 = 19.9: 0.1: 44.6: 35.4 mol%) in zeolite 13X bed are compared by the extended TD Langmuir model and extended TD DSL model. Since the AREs of the TD DSL model are smaller than that of the TD Langmuir, the breakthrough curves of multi-component gas simulated by the TD DSL model are better than that of TD Langmuir. These results can contribute to relevant researchers choosing suitable adsorption isotherm models for numerical simulations of adsorption processes.

Epidemic dynamics in census-calibrated modular contact network.

Network-based models are apt for understanding epidemic dynamics due to their inherent ability to model the heterogeneity of interactions in the contemporary world of intense human connectivity. We propose a framework to create a wire-frame that mimics the social contact network of the population in a geography by lacing it with demographic information. The framework results in a modular network with small-world topology that accommodates density variations and emulates human interactions in family, social, and work spaces. When loaded with suitable economic, social, and urban data shaping patterns of human connectance, the network emerges as a potent decision-making instrument for urban planners, demographers, and social scientists. We employ synthetic networks to experiment in a controlled environment and study the impact of zoning, density variations, and population mobility on the epidemic variables using a variant of the SEIR model. Our results reveal that these demographic factors have a characteristic influence on social contact patterns, manifesting as distinct epidemic dynamics. Subsequently, we present a real-world COVID-19 case study for three Indian states by creating corresponding surrogate social contact networks using available census data. The case study validates that the demography-laced modular contact network reduces errors in the estimates of epidemic variables.

Efficient anomaly detection through surrogate neural networks

Efficient anomaly detection through surrogate neural networks

A 3D-1D coupled blood flow and oxygen transport model to generate microvascular networks.

In this work, we introduce an algorithmic approach to generate microvascular networks starting from larger vessels that can be reconstructed without noticeable segmentation errors. Contrary to larger vessels, the reconstruction of fine-scale components of microvascular networks shows significant segmentation errors, and an accurate mapping is time and cost intense. Thus there is a need for fast and reliable reconstruction algorithms yielding surrogate networks having similar stochastic properties as the original ones. The microvascular networks are constructed in a marching way by adding vessels to the outlets of the vascular tree from the previous step. To optimise the structure of the vascular trees, we use Murray's law to determine the radii of the vessels and bifurcation angles. In each step, we compute the local gradient of the partial pressure of oxygen and adapt the orientation of the new vessels to this gradient. At the same time, we use the partial pressure of oxygen to check whether the considered tissue block is supplied sufficiently with oxygen. Computing the partial pressure of oxygen, we use a 3D-1D coupled model for blood flow and oxygen transport. To decrease the complexity of a fully coupled 3D model, we reduce the blood vessel network to a 1D graph structure and use a bi-directional coupling with the tissue which is described by a 3D homogeneous porous medium. The resulting surrogate networks are analysed with respect to morphological and physiological aspects.

Modeling time-varying brain networks with a self-tuning optimized Kalman filter.

Brain networks are complex dynamical systems in which directed interactions between different areas evolve at the sub-second scale of sensory, cognitive and motor processes. Due to the highly non-stationary nature of neural signals and their unknown noise components, however, modeling dynamic brain networks has remained one of the major challenges in contemporary neuroscience. Here, we present a new algorithm based on an innovative formulation of the Kalman filter that is optimized for tracking rapidly evolving patterns of directed functional connectivity under unknown noise conditions. The Self-Tuning Optimized Kalman filter (STOK) is a novel adaptive filter that embeds a self-tuning memory decay and a recursive regularization to guarantee high network tracking accuracy, temporal precision and robustness to noise. To validate the proposed algorithm, we performed an extensive comparison against the classical Kalman filter, in both realistic surrogate networks and real electroencephalography (EEG) data. In both simulations and real data, we show that the STOK filter estimates time-frequency patterns of directed connectivity with significantly superior performance. The advantages of the STOK filter were even clearer in real EEG data, where the algorithm recovered latent structures of dynamic connectivity from epicranial EEG recordings in rats and human visual evoked potentials, in excellent agreement with known physiology. These results establish the STOK filter as a powerful tool for modeling dynamic network structures in biological systems, with the potential to yield new insights into the rapid evolution of network states from which brain functions emerge.

DEMOCRATISING DESIGN THROUGH SURROGATE MODEL CONVOLUTIONAL NEURAL NETWORKS OF COMPUTER AIDED DESIGN REPOSITORIES

AbstractThe capability to manufacture at home is continually increasing with technologies, such as 3D printing. However, the ability to design products suitable for manufacture and use remains a highly-skilled and knowledge intensive activity. This has led to ‘content creators’ providing vast repositories of manufacturable products for society, however challenges remain in the search & retrieval of models. This paper presents the surrogate model convolutional neural networks approach to search and retrieve CAD models by mapping them directly to their real-world photographed counterparts.

2481. Comparing inter-hospital patient movement patterns to better understand mechanisms for regional dissemination of carbapenem-resistant Enterobacteriaceae

BackgroundUnderstanding inter-hospital movement of patients provides insight into regional transmission of multidrug-resistant organisms (MDROs) that can guide containment efforts. Movement of general patient populations are often used for this purpose, but movement of the specific patient population of MDRO carriers may be more useful. We sought to compare movement of CRE patients with that of other patient populations to explore whether CRE carriers move differently, and if so, to determine whether administrative data can be used to identify patient populations with transfer patterns that mimic CRE patients.MethodsWe used New York’s Statewide Planning and Research Cooperative System (SPARCS), to create a patient network of all acute care hospital encounters (“overall hospital population”) during 2013–2015. We identified the subset of CRE cases in the network by linking the SPARCS data to CRE cases reported to the National Healthcare Safety Network in 2014, matching on admission date, date of birth, gender, and facility. We described patient characteristics and movement patterns across 3 cohorts: (1) CRE cases, (2) overall hospital population, (3) CRE surrogate (patients clinically similar to CRE cases based on length of stay [LOS] ≥14 days and Clinical Classification Software [CCS] category of sepsis plus at least one of the following additional CCS categories: adult respiratory failure, acute renal failure, procedure complication or device complication). Correlations between cohorts were calculated using patient transfer matrices to determine similarities between the networks.ResultsThe average LOS for CRE cases was 25× higher than the overall hospital population (31.4 vs. 1.3 days, Figure 1a), and CRE cases were more likely to die or be discharged to a skilled nursing facility (Figure 1b). CRE movement networks were only moderately correlated with the overall hospital population (R2 = 0.51); there was higher correlation between CRE case and CRE surrogate networks (R2 = 0.73).ConclusionCRE patients have different healthcare experiences in the hospital and between hospitals in New York compared with the overall hospital population. The CRE surrogate cohort transfer patterns were more similar, and could be used to understand CRE patient movement in the absence of CRE culture data. Disclosures All authors: No reported disclosures.

Randomization algorithms for large sparse networks.

In many domains it is necessary to generate surrogate networks, e.g., for hypothesis testing of different properties of a network. Generating surrogate networks typically requires that different properties of the network are preserved, e.g., edges may not be added or deleted and edge weights may be restricted to certain intervals. In this paper we present an efficient property-preserving Markov chain Monte Carlo method termed CycleSampler for generating surrogate networks in which (1) edge weights are constrained to intervals and vertex strengths are preserved exactly, and (2) edge and vertex strengths are both constrained to intervals. These two types of constraints cover a wide variety of practical use cases. The method is applicable to both undirected and directed graphs. We empirically demonstrate the efficiency of the CycleSampler method on real-world data sets. We provide an implementation of CycleSampler in R, with parts implemented in C.

Estimating the outcome of spreading processes on networks with incomplete information: A dimensionality reduction approach.

Recent advances in data collection have facilitated the access to time-resolved human proximity data that can conveniently be represented as temporal networks of contacts between individuals. While the structural and dynamical information revealed by this type of data is fundamental to investigate how information or diseases propagate in a population, data often suffer from incompleteness, which possibly leads to biased estimations in data-driven models. A major challenge is thus to estimate the outcome of spreading processes occurring on temporal networks built from partial information. To cope with this problem, we devise an approach based on non-negative tensor factorization, a dimensionality reduction technique from multilinear algebra. The key idea is to learn a low-dimensional representation of the temporal network built from partial information and to use it to construct a surrogate network similar to the complete original network. To test our method, we consider several human-proximity networks, on which we perform resampling experiments to simulate a loss of data. Using our approach on the resulting partial networks, we build a surrogate version of the complete network for each. We then compare the outcome of a spreading process on the complete networks (nonaltered by a loss of data) and on the surrogate networks. We observe that the epidemic sizes obtained using the surrogate networks are in good agreement with those measured on the complete networks. Finally, we propose an extension of our framework that can leverage additional data, when available, to improve the surrogate network when the data loss is particularly large.

Can co-location be used as a proxy for face-to-face contacts?

Technological advances have led to a strong increase in the number of data collection efforts aimed at measuring co-presence of individuals at different spatial resolutions. It is however unclear how much co-presence data can inform us on actual face-to-face contacts, of particular interest to study the structure of a population in social groups or for use in data-driven models of information or epidemic spreading processes. Here, we address this issue by leveraging data sets containing high resolution face-to-face contacts as well as a coarser spatial localisation of individuals, both temporally resolved, in various contexts. The co-presence and the face-to-face contact temporal networks share a number of structural and statistical features, but the former is (by definition) much denser than the latter. We thus consider several down-sampling methods that generate surrogate contact networks from the co-presence signal and compare them with the real face-to-face data. We show that these surrogate networks reproduce some features of the real data but are only partially able to identify the most central nodes of the face-to-face network. We then address the issue of using such down-sampled co-presence data in data-driven simulations of epidemic processes, and in identifying efficient containment strategies. We show that the performance of the various sampling methods strongly varies depending on context. We discuss the consequences of our results with respect to data collection strategies and methodologies.

Functional complexity emerging from anatomical constraints in the brain: the significance of network modularity and rich-clubs.

The large-scale structural ingredients of the brain and neural connectomes have been identified in recent years. These are, similar to the features found in many other real networks: the arrangement of brain regions into modules and the presence of highly connected regions (hubs) forming rich-clubs. Here, we examine how modules and hubs shape the collective dynamics on networks and we find that both ingredients lead to the emergence of complex dynamics. Comparing the connectomes of C. elegans, cats, macaques and humans to surrogate networks in which either modules or hubs are destroyed, we find that functional complexity always decreases in the perturbed networks. A comparison between simulated and empirically obtained resting-state functional connectivity indicates that the human brain, at rest, lies in a dynamical state that reflects the largest complexity its anatomical connectome can host. Last, we generalise the topology of neural connectomes into a new hierarchical network model that successfully combines modular organisation with rich-club forming hubs. This is achieved by centralising the cross-modular connections through a preferential attachment rule. Our network model hosts more complex dynamics than other hierarchical models widely used as benchmarks.

Activation of surrogate death receptor signaling triggers peroxynitrite-dependent execution of cisplatin-resistant cancer cells.

Platinum-based drugs remain as the cornerstone of cancer chemotherapy; however, development of multidrug resistance presents a therapeutic challenge. This study aims at understanding the molecular mechanisms underlying resistance to cisplatin and unraveling surrogate signaling networks that could revert sensitivity to apoptosis stimuli. We made use of three different sets of cell lines, A549 and H2030 non-small-cell lung cancer (NSCLC) and A2780 ovarian cancer cells and their cisplatin-resistant variants. Here we report that cisplatin-resistant cell lines displayed a multidrug-resistant phenotype. Changes in mitochondrial metabolism and defective mitochondrial signaling were unraveled in the resistant cells. More interestingly, a marked increase in sensitivity of the resistant cells to death receptor-induced apoptosis, in particular TRAIL (TNF-related apoptosis-inducing ligand)-mediated execution, was observed. Although this was not associated with an increase in gene transcription, a significant increase in the localization of TRAIL death receptor, DR4, to the lipid raft subdomains of plasma membrane was detected in the resistant variants. Furthermore, exposure of cisplatin-resistant cells to TRAIL resulted in upregulation of inducible nitric oxide synthase (iNOS) and increase in nitric oxide (NO) production that triggered the generation of peroxynitrite (ONOO−). Scavenging ONOO− rescued cells from TRAIL-induced apoptosis, thereby suggesting a critical role of ONOO− in TRAIL-induced execution of cisplatin-resistant cells. Notably, preincubation of cells with TRAIL restored sensitivity of resistant cells to cisplatin. These data provide compelling evidence for employing strategies to trigger death receptor signaling as a second-line treatment for cisplatin-resistant cancers.

Influence of Wiring Cost on the Large-Scale Architecture of Human Cortical Connectivity

In the past two decades some fundamental properties of cortical connectivity have been discovered: small-world structure, pronounced hierarchical and modular organisation, and strong core and rich-club structures. A common assumption when interpreting results of this kind is that the observed structural properties are present to enable the brain's function. However, the brain is also embedded into the limited space of the skull and its wiring has associated developmental and metabolic costs. These basic physical and economic aspects place separate, often conflicting, constraints on the brain's connectivity, which must be characterized in order to understand the true relationship between brain structure and function. To address this challenge, here we ask which, and to what extent, aspects of the structural organisation of the brain are conserved if we preserve specific spatial and topological properties of the brain but otherwise randomise its connectivity. We perform a comparative analysis of a connectivity map of the cortical connectome both on high- and low-resolutions utilising three different types of surrogate networks: spatially unconstrained (‘random’), connection length preserving (‘spatial’), and connection length optimised (‘reduced’) surrogates. We find that unconstrained randomisation markedly diminishes all investigated architectural properties of cortical connectivity. By contrast, spatial and reduced surrogates largely preserve most properties and, interestingly, often more so in the reduced surrogates. Specifically, our results suggest that the cortical network is less tightly integrated than its spatial constraints would allow, but more strongly segregated than its spatial constraints would necessitate. We additionally find that hierarchical organisation and rich-club structure of the cortical connectivity are largely preserved in spatial and reduced surrogates and hence may be partially attributable to cortical wiring constraints. In contrast, the high modularity and strong s-core of the high-resolution cortical network are significantly stronger than in the surrogates, underlining their potential functional relevance in the brain.

Communication efficiency and congestion of signal traffic in large-scale brain networks.

The complex connectivity of the cerebral cortex suggests that inter-regional communication is a primary function. Using computational modeling, we show that anatomical connectivity may be a major determinant for global information flow in brain networks. A macaque brain network was implemented as a communication network in which signal units flowed between grey matter nodes along white matter paths. Compared to degree-matched surrogate networks, information flow on the macaque brain network was characterized by higher loss rates, faster transit times and lower throughput, suggesting that neural connectivity may be optimized for speed rather than fidelity. Much of global communication was mediated by a “rich club” of hub regions: a sub-graph comprised of high-degree nodes that are more densely interconnected with each other than predicted by chance. First, macaque communication patterns most closely resembled those observed for a synthetic rich club network, but were less similar to those seen in a synthetic small world network, suggesting that the former is a more fundamental feature of brain network topology. Second, rich club regions attracted the most signal traffic and likewise, connections between rich club regions carried more traffic than connections between non-rich club regions. Third, a number of rich club regions were significantly under-congested, suggesting that macaque connectivity actively shapes information flow, funneling traffic towards some nodes and away from others. Together, our results indicate a critical role of the rich club of hub nodes in dynamic aspects of global brain communication.

Boundary effects in network measures of spatially embedded networks

In studies of spatially confined networks, network measures can lead to false conclusions since most measures are boundary affected. This is especially the case if boundaries are artificial and not inherent in the underlying system of interest (e.g., borders of countries). An analytical estimation of boundary effects is not trivial due to the complexity of measures. The straightforward approach we propose here is to use surrogate networks that provide estimates of boundary effects in graph statistics. This is achieved by using spatially embedded random networks as surrogates that have approximately the same link probability as a function of spatial link lengths. The potential of our approach is demonstrated for an analysis of spatial patterns in characteristics of regional climate networks. As an example networks derived from daily rainfall data and restricted to the region of Germany are considered.

Building Smaller Sized Surrogate Models of Complex Bipartite Networks Based on Degree Distributions

This paper presents an approach for generating surrogate bipartite networks with varying sizes based on degree distributions of given bipartite networks. The resulting surrogate networks can be used for problems such as design of algorithms for similarity search, community detection and clustering, and recommender systems. The primary advantage of using smaller surrogate networks over original large-scale networks is the reduction in associated computational expense. Degree distribution is chosen because of its widespread acceptance, simplicity, and prior literature suggesting its ability to better capture large-scale network properties. The approach is illustrated using a bipartite network from an open-source software development repository. The network consists of nodes representing people and projects, and edges representing people working on different projects. A comparison between the surrogate networks and the original networks is presented. The results show that the resized networks obtained using the proposed approach can be used to match the original degree distribution. A comparison of seven other network characteristics is also provided.

Surrogate-assisted analysis of weighted functional brain networks

Graph-theoretical analyses of complex brain networks is a rapidly evolving field with a strong impact for neuroscientific and related clinical research. Due to a number of confounding variables, however, a reliable and meaningful characterization of particularly functional brain networks is a major challenge. Addressing this problem, we present an analysis approach for weighted networks that makes use of surrogate networks with preserved edge weights or vertex strengths. We first investigate whether characteristics of weighted networks are influenced by trivial properties of the edge weights or vertex strengths (e.g., their standard deviations). If so, these influences are then effectively segregated with an appropriate surrogate normalization of the respective network characteristic. We demonstrate this approach by re-examining, in a time-resolved manner, weighted functional brain networks of epilepsy patients and control subjects derived from simultaneous EEG/MEG recordings during different behavioral states. We show that this surrogate-assisted analysis approach reveals complementary information about these networks, can aid with their interpretation, and thus can prevent deriving inappropriate conclusions.