Network Model Selection Research Articles

A statistical modelling approach to feedforward neural network model selection

Feedforward neural networks (FNNs) can be viewed as non-linear regression models, where covariates enter the model through a combination of weighted summations and non-linear functions. Although these models have some similarities to the approaches used within statistical modelling, the majority of neural network research has been conducted outside of the field of statistics. This has resulted in a lack of statistically based methodology, and, in particular, there has been little emphasis on model parsimony. Determining the input layer structure is analogous to variable selection, while the structure for the hidden layer relates to model complexity. In practice, neural network model selection is often carried out by comparing models using out-of-sample performance. However, in contrast, the construction of an associated likelihood function opens the door to information-criteria-based variable and architecture selection. A novel model selection method, which performs both input- and hidden-node selection, is proposed using the Bayesian information criterion (BIC) for FNNs. The choice of BIC over out-of-sample performance as the model selection objective function leads to an increased probability of recovering the true model, while parsimoniously achieving favourable out-of-sample performance. Simulation studies are used to evaluate and justify the proposed method, and applications on real data are investigated.

Water vapor content prediction based on neural network model selection and optimal fusion

Accurately predicting the evolution of water vapor content has significant practical engineering implications for accurately determining the timing of artificial rain enhancement operations. However, the nonlinear, unstable, and chaotic nature of water vapor content presents considerable challenges to its precise prediction. To address this, our research proposes a combined prediction model, based on a neural network model selection mechanism and small sample water vapor content data observed by microwave radiometers in the Zhengzhou area. This combined model utilizes seven neural network models as candidate models and dynamically selects the optimal three models as the central prediction components via evaluation indicators. Additionally, we employ the maximum information coefficient method to preserve the optimal input features and use a time-varying rate wave mode decomposition algorithm as the front-end processing module. This module processes the original water vapor content sequence and extracts relevant latent information. Finally, we incorporate the African vulture intelligent optimization algorithm as the backend processing component to optimize the weighted combination coefficients, achieving the integration of prediction results. Experimental results reveal that the Mean Absolute Percentage Errors (MAPE) of the proposed model on three types of water vapor content datasets are 0.208%, 0.876%, and 0.369%, respectively. This suggests a predictive performance improvement of at least 10% compared to existing models. The prediction model proposed in this study not only establishes an accurate and reliable mechanism for predicting water vapor content but also offers more comprehensive decision support information for weather modification operations.

Deep Learning-based Agricultural Pest and Disease Recognition

With the rapid development of artificial intelligence, deep learning technology has penetrated into a variety of different fields. Agricultural researchers have used deep learning's strong image and data processing capabilities to apply it to agricultural production, especially in the identification of agricultural pests and diseases, which has achieved remarkable results. The use of deep learning-based agricultural pest and disease identification technology can quickly and non-destructively identify crops in various complex environments, with high accuracy and good stability, so that farmers can quickly take effective control measures to minimize the loss of crops. This paper first explains the significance of deep learning-based agricultural pest identification technology to agricultural development, and makes a detailed analysis of the advantages and disadvantages of various pest identification technologies and their performance in pest and disease identification. Secondly, various key technologies of deep learning are introduced, including the concepts and applications of various core technologies such as data source, data preprocessing, data augmentation, network model selection, and transfer learning. Finally, based on the above research results, the problems encountered by deep learning in the identification of agricultural pests and diseases are analyzed, and its future development direction in the identification of pests and diseases is discussed.

Accelerating Adversarially Robust Model Selection for Deep Neural Networks via Racing

Recent research has introduced several approaches to formally verify the robustness of neural network models against perturbations in their inputs, such as the ones that occur in adversarial attacks. At the same time, this particular verification task is known to be computationally challenging. More specifically, assessing the robustness of a neural network against input perturbations can easily take several hours of compute time per input vector, even when using state-of-the-art verification approaches. In light of this, it becomes challenging to select from a given set of neural network models the one that is best in terms of robust accuracy, i.e., the fraction of instances for which the model is known to be robust against adversarial perturbations, especially when given limited computing resources. To tackle this problem, we propose a racing method specifically adapted to the domain of robustness verification. This racing method utilises Delta-values, which can be seen as an efficiently computable proxy for the distance of a given input to a neural network model to the decision boundary. We present statistical evidence indicating significant differences in the empirical cumulative distribution between robust and non-robust inputs as a function of Delta-values. Using this information, we show that it is possible to reliably expose vulnerabilities in the model with relatively few input iterations. Overall, when applied to selecting the most robust network from sets of 31 MNIST and 27 CIFAR-10 networks, our proposed method achieves speedups of a factor of 108 and 42, respectively, in terms of cumulative running time compared to standard local robustness verification on the complete testing sets.

Automated Lung and Colon Cancer Classification Using Histopathological Images.

Cancer is the leading cause of mortality in the world. And among all cancers lung and colon cancers are 2 of the most common causes of death and morbidity. The aim of this study was to develop an automated lung and colon cancer classification system using histopathological images. An automated lung and colon classification system was developed using histopathological images from the LC25000 dataset. The algorithm development included data splitting, deep neural network model selection, on the fly image augmentation, training and validation. The core of the algorithm was a Swin Transform V2 model, and 5-fold cross validation was used to evaluate model performance. The model performance was evaluated using Accuracy, Kappa, confusion matrix, precision, recall, and F1. Extensive experiments were conducted to compare the performances of different neural networks including both mainstream convolutional neural networks and vision transformers. The Swin Transform V2 model achieved a 1 (100%) on all metrics, which is the first single model to obtain perfect results on this dataset. The Swin Transformer V2 model has the potential to be used to assist pathologists in classifying lung and colon cancers using histopathology images.

AutoKeras for Fake News Identification in Arabic: Leveraging Deep Learning with an Extensive Dataset

Social media and the World Wide Web have led to a worrying rise in spreading false information, which presents a significant worldwide issue. Identifying and preventing false information is crucial in promoting an informed and knowledgeable society. The identification of false information, specifically in the Arabic dialect, presents inherent difficulties due to its diverse characteristics and linguistic intricacies. This study implements AutoKeras, a deep learning-based machine learning framework. Using advanced optimization techniques, the neural network architecture search, hyperparameter adjustments, and model selection can all be automated in AutoKeras. Therefore, it is suitable for our fake news detection task. The methodology employs proficient deep learning algorithms and natural language processing methods to acquire distinct characteristics that enable accurate differentiation between genuine and fake news. The present study uses various sources, including news websites, social media platforms, and blogs, to construct the dataset. The AutoKeras-based approach is superior to multiple state-of-the-art approaches to detecting fabricated news in Arabic, as evidenced by the experimental results. The suggested method outperforms 93.2% accuracy in identifying fake news, demonstrating its superior efficacy. This demonstrates the great promise of the deep learning-based Auto model for detecting false information.

Cost-based Feature Selection for Network Model Choice

Selecting a small set of informative features from a large number of possibly noisy candidates is a challenging problem with many applications in machine learning and approximate Bayesian computation. In practice, the cost of computing informative features also needs to be considered. This is particularly important for networks because the computational costs of individual features can span several orders of magnitude. We addressed this issue for the network model selection problem using two approaches. First, we adapted nine feature selection methods to account for the cost of features. We show for two classes of network models that the cost can be reduced by two orders of magnitude without considerably affecting classification accuracy (proportion of correctly identified models). Second, we selected features using pilot simulations with smaller networks. This approach reduced the computational cost by a factor of 50 without affecting classification accuracy. To demonstrate the utility of our approach, we applied it to three different yeast protein interaction networks and identified the best-fitting duplication divergence model. Supplementary materials, including computer code to reproduce our results, are available online.

Hierarchical Model Selection for Graph Neural Networks

Node classification on graph data is a major problem in machine learning, and various graph neural networks (GNNs) have been proposed. Variants of GNNs such as H2GCN and CPF outperform graph convolutional networks (GCNs) by improving on the weaknesses of the traditional GNN. However, there are some graph data which these GNN variants fail to perform well than other GNNs in the node classification task. This is because H2GCN has similar feature values on graph data with high average degree, and CPF gives rise to a problem about label-propagation suitability. Accordingly, we propose a hierarchical model selection framework (HMSF) that selects an appropriate GNN model to predict the class of nodes for each graph data. HMSF uses average degree and edge homophily ratio as indicators to decide the useful model based on our analyses. In the experiment, we show that the model selected by our HMSF achieves high performance on node classification for various types of graph data.

Privacy-Preserving Mobility Model and Optimization-Based Advanced Cluster Head Selection (P2O-ACH) for Vehicular Ad Hoc Networks

In vehicular ad hoc networks (VANETs), due to the fast-moving mobile nodes, the topology changes frequently. This dynamically changing topology produces congestion and instability. To overcome this issue, privacy-preserving optimization-based cluster head selection (P2O-ACH) is proposed. One of the major drawbacks analyzed in the earlier cluster-based VANETs is that it creates a maximum number of clusters for communication that leads to an increase in energy consumption which reflects in a degradation of the performance. In this paper, enhanced rider optimization algorithm (ROA)-based CH selection is performed and that optimally selects the CH so that effective clusters are created. By analyzing this, the behavior of the bypass rider’s CH is chosen, and this forms the optimized clusters, and during the process of transmission, privacy-preserving mobility patterns are used to secure the network from all kinds of malfunctions which are performed by the new vehicle blending and migration process. The proposed P2O-ACH is simulated using NS-2, and for performance analysis, two scenarios are taken, which contain a varying number of vehicles and varying speeds. For a varying number of vehicles and speeds, the considered parameters are energy efficiency, energy consumption, network lifetime, packet delivery ratio, packet loss, network latency, network throughput, and routing overhead. From the results, it is understood that the proposed method performed better when compared with earlier work, such as GWO-CH, ACO-SCRS, and QMM-VANET.

Methodological evaluation of individual cognitive prediction based on the brain white matter structural connectome.

An emerging trend is to use regression‐based machine learning approaches to predict cognitive functions at the individual level from neuroimaging data. However, individual prediction models are inherently influenced by the vast options for network construction and model selection in machine learning pipelines. In particular, the brain white matter (WM) structural connectome lacks a systematic evaluation of the effects of different options in the pipeline on predictive performance. Here, we focused on the methodological evaluation of brain structural connectome‐based predictions. For network construction, we considered two parcellation schemes for defining nodes and seven strategies for defining edges. For the regression algorithms, we used eight regression models. Four cognitive domains and brain age were targeted as predictive tasks based on two independent datasets (Beijing Aging Brain Rejuvenation Initiative [BABRI]: 633 healthy older adults; Human Connectome Projects in Aging [HCP‐A]: 560 healthy older adults). Based on the results, the WM structural connectome provided a satisfying predictive ability for individual age and cognitive functions, especially for executive function and attention. Second, different parcellation schemes induce a significant difference in predictive performance. Third, prediction results from different data sets showed that dMRI with distinct acquisition parameters may plausibly result in a preference for proper fiber reconstruction algorithms and different weighting options. Finally, deep learning and Elastic‐Net models are more accurate and robust in connectome‐based predictions. Together, significant effects of different options in WM network construction and regression algorithms on the predictive performances are identified in this study, which may provide important references and guidelines to select suitable options for future studies in this field.

Multi-objective neural network model selection with a graph-based large margin approach

This work presents a new decision-making strategy for multi-objective learning problem of artificial neural networks (ANN). The proposed decision-maker searches for the solution that minimizes a margin-based validation error amongst Pareto set solutions. The proposal is based on a geometric approximation to find the large margin (distance) of separation among the classes. Several benchmarks commonly available in the literature were used for testing. The obtained results showed that the proposal is more efficient in controlling the generalization capacity of neural models than other learning machines. It yields smooth (noise robustness) and well-fitted models straightforwardly, i.e., without the necessity of parameter set definition in advance or validation data use, as often required by learning machines.

A Single Stream Modified MobileNet V2 and Whale Controlled Entropy Based Optimization Framework for Citrus Fruit Diseases Recognition

Fruit disease recognition is quickly becoming a hot topic in the field of computer vision. The presence of plant diseases not only reduces fruit production but also causes a significant loss to the national economy. Citrus fruits help to strengthen the immune system, allowing it to fight off diseases such as COVID-19. Manual inspection of fruit diseases with the naked eye takes time and is difficult; therefore, a computer based method is always required for accurate recognition of plant diseases. Several deep learning techniques for recognizing citrus fruit diseases have been introduced in the literature. Existing techniques had several issues, including redundant features, convolutional neural network (CNN) model selection, low contrast images, and long computational times. In this paper, a single stream convolutional neural network architecture is proposed for recognizing citrus fruit diseases. In the first step, data augmentation is performed using four contrast enhancement operations: shadow removal, adjusting pixel intensity, improving brightness, and improving local contrast. The MobileNet-V2 CNN model is selected and fine-tuned in the second step. Using the transfer learning process, the fine-tuned model is trained on the augmented citrus dataset. The newly trained model is used for deep feature extraction; however, analysis shows that the extracted deep features contain little redundant information. As a result, an improved Whale Optimization Algorithm (IWOA) is used in the third step. The best features are then classified using machine learning classifiers in the final step. The augmented citrus fruits, leaves, and hybrid dataset were used in the experimental process and achieved an accuracy of 99.4, 99.5, and 99.7%. When compared to existing techniques, the proposed architecture outperformed them in terms of accuracy and time.

Deep learning based network similarity for model selection

Capturing data in the form of networks is becoming an increasingly popular approach for modeling, analyzing and visualising complex phenomena, to understand the important properties of the underlying complex processes. Access to many large-scale network datasets is restricted due to the privacy and security concerns. Also for several applications (such as functional connectivity networks), generating large scale real data is expensive. For these reasons, there is a growing need for advanced mathematical and statistical models (also called generative models) that can account for the structure of these large-scale networks, without having to materialize them in the real world. The objective is to provide a comprehensible description of the network properties and to be able to infer previously unobserved properties. Various models have been developed by researchers, which generate synthetic networks that adhere to the structural properties of real networks. However, the selection of the appropriate generative model for a given real-world network remains an important challenge. In this paper, we investigate this problem and provide a novel technique (named as TripletFit) for model selection (or network classification) and estimation of structural similarities of the complex networks. The goal of network model selection is to select a generative model that is able to generate a structurally similar synthetic network for a given real-world (target) network. We consider six outstanding generative models as the candidate models. The existing model selection methods mostly suffer from sensitivity to network perturbations, dependency on the size of the networks, and low accuracy. To overcome these limitations, we considered a broad array of network features, with the aim of representing different structural aspects of the network and employed deep learning techniques such as deep triplet network architecture and simple feed-forward network for model selection and estimation of structural similarities of the complex networks. Our proposed method, outperforms existing methods with respect to accuracy, noise-tolerance, and size independence on a number of gold standard data set used in previous studies.

Training Image Optimization Method Based on Convolutional Neural Network and Its Application in Discrete Fracture Network Model Selection

Abstract Training image (TI) is important for multipoint statistics simulation method (MPS), since it captures the spatial geological pattern of target reservoir to be modeled. Generally, one optimal TI is selected before applying MPS by evaluating the similarities between many TIs and the well interpretations of target reservoir. In this paper, we propose a new training image optimization approach based on the convolutional neural network (CNN). First, candidate TIs were randomly sampled several times to obtain the sample dataset. Then, the CNN was used to conduct transfer learning for all samples, and finally, the optimal TI of the conditioning well data is selected through the trained CNN model. By taking advantage of the strong learning ability of CNN in image feature recognition, the proposed method can automatically identify differences in spatial features between the conditioning well data and the samples of the training image. Hence, it effectively resolves the difficulty of spatial matching between discrete datapoints and grid structures. We demonstrated the applicability of our model via 2D and 3D training image selection examples. The proposed methods effectively selected the appropriate TI, and then the pretreatment techniques for improving the accuracy of continuous TI selection were achieved. Moreover, the proposed method was successfully applied to training image selection of a discrete fracture network model. Finally, sensitivity analysis was carried out to show that sufficient conditioning data volume can reduce the uncertainty of the optimization results. By comparing with the improved MDevD method, the advantages of the new method are verified in terms of efficiency and reliability.

Investigation of patient-sharing networks using a Bayesian network model selection approach for congruence class models.

A Bayesian approach to conduct network model selection is presented for a general class of network models referred to as the congruence class models (CCMs). CCMs form a broad class that includes as special cases several common network models, such as the Erdős-Rényi-Gilbert model, stochastic block model, and many exponential random graph models. Due to the range of models that can be specified as CCMs, our proposed method is better able to select models consistent with generative mechanisms associated with observed networks than are current approaches. In addition, our approach allows for incorporation of prior information. We illustrate the use of this approach to select among several different proposed mechanisms for the structure of patient-sharing networks; such networks have been found to be associated with the cost and quality of medical care. We found evidence in support of heterogeneity in sociality but not selective mixing by provider type or degree.

Inference, Model Selection, and the Combinatorics of Growing Trees.

One can often make inferences about a growing network from its current state alone. For example, it is generally possible to determine how a network changed over time or pick among plausible mechanisms explaining its growth. In practice, however, the extent to which such problems can be solved is limited by existing techniques, which are often inexact, inefficient, or both. In this Letter, we derive exact and efficient inference methods for growing trees and demonstrate them in a series of applications: network interpolation, history reconstruction, model fitting, and model selection.

Network cross-validation by edge sampling

Summary While many statistical models and methods are now available for network analysis, resampling of network data remains a challenging problem. Cross-validation is a useful general tool for model selection and parameter tuning, but it is not directly applicable to networks since splitting network nodes into groups requires deleting edges and destroys some of the network structure. In this paper we propose a new network resampling strategy, based on splitting node pairs rather than nodes, that is applicable to cross-validation for a wide range of network model selection tasks. We provide theoretical justification for our method in a general setting and examples of how the method can be used in specific network model selection and parameter tuning tasks. Numerical results on simulated networks and on a statisticians’ citation network show that the proposed cross-validation approach works well for model selection.

Genetic assignment of individuals to source populations using network estimation tools

Abstract Dispersal, the movement of individuals between populations, is crucial in many ecological and genetic processes. However, direct identification of dispersing individuals is difficult or impossible in natural populations. By using genetic assignment methods, individuals with unknown genetic origin can be assigned to source populations. This knowledge is necessary in studying many key questions in ecology, evolution and conservation. We introduce a network‐based tool BONE (Baseline Oriented Network Estimation) for genetic population assignment, which borrows concepts from undirected graph inference. In particular, we use sparse multinomial Least Absolute Shrinkage and Selection Operator (LASSO) regression to estimate probability of the origin of all mixture individuals and their mixture proportions without tedious selection of the LASSO tuning parameter. We compare BONE with three genetic assignment methods implemented in R packages radmixture, assignPOP and RUBIAS. Probability of the origin and mixture proportion estimates of both simulated and real data (an insular house sparrow metapopulation and Chinook salmon populations) given by BONE are competitive or superior compared to other assignment methods. Our examples illustrate how the network estimation method adapts to population assignment, combining the efficiency and attractive properties of sparse network representation and model selection properties of the L1 regularization. As far as we know, this is the first approach showing how one can use network tools for genetic identification of individuals' source populations. BONE is aimed at any researcher performing genetic assignment and trying to infer the genetic population structure. Compared to other methods, our approach also identifies outlying mixture individuals that could originate outside of the baseline populations. BONE is a freely available R package under the GPL licence and can be downloaded at GitHub. In addition to the R package, a tutorial for BONE is available at https://github.com/markkukuismin/BONE/.

Flexible model selection for mechanistic network models.

Network models are applied across many domains where data can be represented as a network. Two prominent paradigms for modelling networks are statistical models (probabilistic models for the observed network) and mechanistic models (models for network growth and/or evolution). Mechanistic models are better suited for incorporating domain knowledge, to study effects of interventions (such as changes to specific mechanisms) and to forward simulate, but they typically have intractable likelihoods. As such, and in a stark contrast to statistical models, there is a relative dearth of research on model selection for such models despite the otherwise large body of extant work. In this article, we propose a simulator-based procedure for mechanistic network model selection that borrows aspects from Approximate Bayesian Computation along with a means to quantify the uncertainty in the selected model. To select the most suitable network model, we consider and assess the performance of several learning algorithms, most notably the so-called Super Learner, which makes our framework less sensitive to the choice of a particular learning algorithm. Our approach takes advantage of the ease to forward simulate from mechanistic network models to circumvent their intractable likelihoods. The overall process is flexible and widely applicable. Our simulation results demonstrate the approach's ability to accurately discriminate between competing mechanistic models. Finally, we showcase our approach with a protein-protein interaction network model from the literature for yeast (Saccharomyces cerevisiae).

Hybrid Stochastic GA-Bayesian Search for Deep Convolutional Neural Network Model Selection.

Hybrid Stochastic GA-Bayesian Search for Deep Convolutional Neural Network Model Selection.