Network Cost Function Research Articles

Noisy Quantum Channel Characterization Using Quantum Neural Networks

Channel noise is considered to be the main obstacle in long-distance quantum communication and distributed quantum networks. Here, employing a quantum neural network, we present an efficient method to study the model and detect the noise of quantum channels. Based on various types of noisy quantum channel models, we construct the architecture of the quantum neural network and the model training process. Finally, we perform experiments to verify the training effectiveness of the scheme, and the results show that the cost function of the quantum neural network could approach above 90% of the channel model.

Joint Resource Scheduling for AMR Navigation Over Wireless Edge Networks

The future of autonomous systems will rely on the usage of wireless time-sensitive networks to connect mobile cyberphysical systems, such as Autonomous Mobile Robots (AMRs), to Edge compute platforms to offload computationally intensive workloads necessary to complete tasks. In the case of AMRs, due to their mobility, the offloading of expensive processes such as localization and tracking methods to the Edge computing infrastructure must also be done over dynamic wireless networks. In larger scale systems, the network and compute resource requirements can quickly become prohibitively large due to network traffic and heavy workloads and tight deadline requirements for proper execution of time-critical tasks. In this paper, we formulate the problem of jointly allocating network and compute resources for time sensitive systems as the state of the wireless channel changes over time. By characterizing a compute model for AMR workloads, we further demonstrate how the network and compute scheduling decisions can be serialized, thus making the optimal scheduling problem significantly more tractable, via the incorporation of a compute-utility aware network cost function. Simulation results of AMR systems in a Wi-Fi network demonstrate substantial gains over baseline scheduling methods in total resource efficiency.

An adaptive fuzzy logic controller for intelligent drying

A systematic approach to the design of an adaptive fuzzy logic controller (AFLC) for intelligent drying with a computer vision system (CVS) in a feedback loop is proposed. Developed AFLC is based on an artificial neural network (ANN), geno-fuzzy algorithm, and multi-objective fuzzy cost function. Fuzzy sets for the moisture content and product quality are automatically generated by using principal component analysis (PCA) and fuzzy clustering. In addition, the concept of fuzzy time is introduced to optimize the duration of each control step. The fuzzy rule base for the controller was constructed through a two-stage process of (i) warming-up based on simulation and optimization (offline) and (ii) fine-tuning during real-time drying (online). The application of AFLC for shrimp drying showed advantages of the unsupervised fuzzy logic control, such as decreased drying time, less quality degradation, and smaller energy consumption.

Optimizing the operator’s economy in multi-tier cellular networks with underlaid D2D communications

In this paper, we consider a three-tier cellular network, consisting of a Macrocell base station, several Femtocell base stations, cellular users, and Device-to-Device (D2D) users. Each user is attached to the Macrocell or one of the Femtocells while D2D users communicate directly. The Macrocell divides the spectrum into several subchannels. The Femtocell users are allowed to communicate over the common spectrum which is allocated to one of the Macrocell users. We consider the scenario in which the Femtocells do not interfere each other while the Macrocell users experience interference from Femtocells and vice versa. Likewise, the Macrocell users would suffer from the interferences caused by underlaid Femtocells and D2D communications in the same spectrum. The average outage probability over the shared sub-channel is investigated assuming the effect of D2D users, and the closed form expressions are obtained in an interference limited scheme. Then, the object function is defined to minimize the operator’s cost function for establishing Femtocell network. The problem is generally non-convex and hence, it is solved through separation to subproblems to derive the near-optimal solutions for the intensity of Femtocells and power allocation. The analytical results are verified by the simulation results, which show that although the D2D and Femtocells improve the network performance by handling the data traffic, increasing the D2D users’ transmission power and the intensity of physical distribution of Femtocells increase the network cost function.

Numerical solution of Generalized Burger–Huxley & Huxley’s equation using Deep Galerkin neural network method

In this paper, a deep learning algorithm based on Deep Galerkin method (DGM) is presented for the approximate solution of the generalized Burgers–Huxley equation (gBHE), and generalized Huxley’s equation (gHE). In this method, a deep neural network (DNN) is used for approximating the solution without generating mesh grid, which satisfies the differential operator, boundary and initial conditions. DNN is trained on randomly selected batches of time and space points, thus helping to avoid forming a mesh. Adam optimizer is used for optimizing the parameters of the DNN. Further, the convergence of the cost function and convergence of the neural network to the exact solution is demonstrated. This method shows very encouraging results which have been compared with recent methods such as: A fourth order improved numerical scheme (FDS4), Adomain-decomposition method (ADM), Modified cubic B-spline differential quadrature method (MCB-DQM), Variational iteration method (VIM), and others.

Emergence of associative learning in a neuromorphic inference network

Objective. In the theoretical framework of predictive coding and active inference, the brain can be viewed as instantiating a rich generative model of the world that predicts incoming sensory data while continuously updating its parameters via minimization of prediction errors. While this theory has been successfully applied to cognitive processes—by modelling the activity of functional neural networks at a mesoscopic scale—the validity of the approach when modelling neurons as an ensemble of inferring agents, in a biologically plausible architecture, remained to be explored. Approach. We modelled a simplified cerebellar circuit with individual neurons acting as Bayesian agents to simulate the classical delayed eyeblink conditioning protocol. Neurons and synapses adjusted their activity to minimize their prediction error, which was used as the network cost function. This cerebellar network was then implemented in hardware by replicating digital neuronal elements via a low-power microcontroller. Main results. Persistent changes of synaptic strength—that mirrored neurophysiological observations—emerged via local (neurocentric) prediction error minimization, leading to the expression of associative learning. The same paradigm was effectively emulated in low-power hardware showing remarkably efficient performance compared to conventional neuromorphic architectures. Significance. These findings show that: (a) an ensemble of free energy minimizing neurons—organized in a biological plausible architecture—can recapitulate functional self-organization observed in nature, such as associative plasticity, and (b) a neuromorphic network of inference units can learn unsupervised tasks without embedding predefined learning rules in the circuit, thus providing a potential avenue to a novel form of brain-inspired artificial intelligence.

Proactive and AoI-Aware Failure Recovery for Stateful NFV-Enabled Zero-Touch 6G Networks: Model-Free DRL Approach

In this paper, we propose a Zero-Touch, deep reinforcement learning (DRL)-based Proactive Failure Recovery framework called ZT-PFR for stateful network function virtualization (NFV)-enabled networks. To this end, we formulate a resource-efficient optimization problem minimizing the network cost function including resource cost and wrong decision penalty. As a solution, we propose state-of-the-art DRL-based methods such as soft-actor-critic (SAC) and proximal-policy-optimization (PPO). In addition, to train and test our DRL agents, we propose a novel impending-failure model. Moreover, to keep network status information at an acceptable freshness level for appropriate decision-making, we apply the concept of age of information to strike a balance between the event and scheduling based monitoring. Several key systems and DRL algorithm design insights for ZT-PFR are drawn from our analysis and simulation results. For example, we use a hybrid neural network, consisting long short-term memory layers in the DRL agents structure, to capture impending-failures time dependency.

Scatter Correction in X-Ray CT by Physics-Inspired Deep Learning

Scatter due to interaction of photons with the imaged object is a fundamental problem in X-ray Computed Tomography (CT). It manifests as various artifacts in the reconstruction, making its abatement or correction critical for image quality. Despite success in specific settings, hardware-based methods require modification in the hardware, or increase in the scan time or dose. This accounts for the great interest in software-based methods, including Monte-Carlo based scatter estimation, analytical-numerical, and kernel-based methods, with data-driven learning-based approaches demonstrated recently. In this work, two novel physics-inspired deep-learning-based methods, PhILSCAT and OV-PhILSCAT, are proposed. The methods estimate and correct for the scatter in the acquired projection measurements. Different to previous works, they incorporate both an initial reconstruction of the object of interest and the scatter-corrupted measurements related to it, and use a deep neural network architecture and cost function, both specifically tailored to the problem. Numerical experiments with data generated by Monte-Carlo simulations of the imaging of phantoms reveal advantages over our implementation of a recent purely projection-domain deep neural network scatter correction method.

A simple model of network formation with competition effects

This paper provides a game-theoretic model of network formation with a continuous effort choice. Efforts are strategic complements for direct neighbors in the network and display global substitution/competition effects. We show that if the parameter governing local strategic complements is larger than the one governing global strategic substitutes, then all pairwise Nash equilibrium networks are nested split graphs. We also consider the problem of a planner, who can choose effort levels and place links according to a network cost function. Again all socially optimal configurations are such that the network is a nested split graph. However, the socially optimal network may be different from equilibrium networks and efficient effort levels do not coincide with Nash equilibrium effort levels. In the presence of strategic substitutes, Nash equilibrium effort levels may be too high or too low relative to efficient effort levels. The relevant applications are crime networks and R&D collaborations among firms, but also interbank lending and trade.

Novel Convolutional Neural Network (NCNN) for the Diagnosis of Bearing Defects in Rotary Machinery

This work presents the development of novel convolutional neural network (NCNN) for effective identification of bearing defects from small samples. For effective feature learning from small training data, cost function of convolution neural network (CNN) is modified by adding additional sparsity cost in the existing cost function. A novel trigonometric cross-entropy function is developed to compute the sparsity cost. The proposed cost function introduces sparsity by avoiding unnecessary activation of neurons in the hidden layers of CNN. For identification of bearing defects from small training samples, NCNN-based transfer learning is applied in the following manner. First, the raw vibration signals as well as envelope signals from source domain machine are obtained. Thereafter, these envelope signals are applied to NCNN for the learning of features from the big training data acquitted from the source domain. After feature learning, knowledge gained from NCNN is transferred to do fine-tuning of NCNN from small training samples of target domain. Thereafter, defect identification is carried out by applying the test data of target domain to fine-tuned NCNN. The experimental result validates that the proposed cross-entropy function introduces sparsity in CNN and, hence, creates an effective deep learning which can even work under a situation when training data are not available in abundant.

Improved CNN for the diagnosis of engine defects of 2-wheeler vehicle using wavelet synchro-squeezed transform (WSST)

In this work, deep learning based diagnostic procedure is developed for the identification of engine defects of 2-wheeler vehicle. The process starts with acquisition of vibration data. Second, time domain signals are converted into angular domain. Third, random distribution of angular domain signals is done to have training and test data. Further, processing of training and test data is carried out using wavelet synchro-squeezed transform (WSST) to form time–frequency images. Then, cost function of convolution neural network (CNN) is modified by introducing a new entropy-based regularization function in the existing cost function which can meaningfully reduce the activation in the hidden layer of CNN so as to make the learning really deep. Thereafter, training of improved CNN is carried out using WSST images of training samples. In the next step, WSST images of test data are applied to tuned CNN for the identification of defects. A comparison of proposed method has been carried by existing deep learning solutions and the method proposed in the state-of-artwork. The comparison shows that the proposed method is at least 3.8 % more superior in terms of accuracy than the existing defect diagnosis methods while diagnosing defects of internal combustion (IC) engine of 2-wheeler vehicle.

Reverse-Engineering Neural Networks to Characterize Their Cost Functions.

This letter considers a class of biologically plausible cost functions for neural networks, where the same cost function is minimized by both neural activity and plasticity. We show that such cost functions can be cast as a variational bound on model evidence under an implicit generative model. Using generative models based on partially observed Markov decision processes (POMDP), we show that neural activity and plasticity perform Bayesian inference and learning, respectively, by maximizing model evidence. Using mathematical and numerical analyses, we establish the formal equivalence between neural network cost functions and variational free energy under some prior beliefs about latent states that generate inputs. These prior beliefs are determined by particular constants (e.g., thresholds) that define the cost function. This means that the Bayes optimal encoding of latent or hidden states is achieved when the network's implicit priors match the process that generates its inputs. This equivalence is potentially important because it suggests that any hyperparameter of a neural network can itself be optimized-by minimization with respect to variational free energy. Furthermore, it enables one to characterize a neural network formally, in terms of its prior beliefs.

When is Selfish Routing Bad? The Price of Anarchy in Light and Heavy Traffic

This paper examines the behavior of the price of anarchy as a function of the traac innow in nonatomic congestion games with multiple origin-destination (O/D) pairs. Empirical studies in real-world networks show that the price of anarchy is close to 1 in both light and heavy traac, thus raising the question: can these observations be justiied theoretically? We rst show that this is not always the case: the price of anarchy may remain a positive distance away from 1 for all values of the traac innow, even in simple three-link networks with a single O/D pair and smooth, convex costs. On the other hand, for a large class of cost functions (including all polynomials), the price of anarchy does converge to 1 in both heavy and light traac, irrespective of the network topology and the number of O/D pairs in the network. We also examine the rate of convergence of the price of anarchy, and we show that it follows a power law whose degree can be computed explicitly when the network's cost functions are polynomials.

High cursive traditional Asian character recognition using integrated adaptive constraints in ensemble of DenseNet and Inception models

In this paper, we propose integrated adaptive sensitivity and robustness terms for the cost function of a convolutional neural network (CNN). The sensitivity term considers the slight variations and high frequency components of the input image samples. It distinguishes between images that look similar but belong to different classes. This regularizer is designed to enhance the between-class distance which is a biological definition for the simple cells of the visual system. On the other hand, the robustness term is used to develop a more stable CNN structure against disturbances and perturbations. The robust term provides better within-class features because it recognizes images that look different but are actually from the same class. The robust term symbolizes the complex cell characteristics of the visual system. The coefficients of the sensitivity and robustness regularization terms are adaptively tuned along with the network parameters using gradient descent. Two optimizers are assigned to tune the parameters: one for tuning the model parameters and the other one to adjust the sensitivity and robustness coefficients. This approach is applied to Korean traditional documents for character classification. The results show better within- and between-class classification ability for highly complex character styles with imbalanced number of samples.

Distributed Signal Processing and Optimization Based on In-Network Subspace Projections

We study distributed optimization and processing of subspace-constrained signals in multi-agent networks with sparse connectivity. We introduce the first optimization framework based on distributed subspace projections, aimed at minimizing a network cost function depending on the specific processing task, while imposing subspace constraints on the final solution. The proposed method hinges on (sub)gradient optimization techniques while leveraging distributed projections as a mechanism to enforce subspace constraints in a cooperative and distributed fashion. Asymptotic convergence rates to optimal solutions of the problem are established under different assumptions (e.g., nondifferentiability, nonconvexity, etc.) on the objective function. We also introduce an extension of the framework that works with constant step-sizes, thus enabling faster convergence to optimal solutions of the optimization problem. Our algorithmic framework is very flexible and can be customized to a variety of problems in distributed signal processing. Finally, numerical tests on synthetic and realistic data illustrate how the proposed methods compare favorably to existing distributed algorithms.

Unsupervised physics-based neural networks for seismic migration

We have developed a novel framework for combining physics-based forward models and neural networks to advance seismic processing and inversion algorithms. Migration is an effective tool in seismic data processing and imaging. Over the years, the scope of these algorithms has broadened; today, migration is a central step in the seismic data processing workflow. However, no single migration technique is suitable for all kinds of data and all styles of acquisition. There is always a compromise on the accuracy, cost, and flexibility of these algorithms. On the other hand, machine-learning algorithms and artificial intelligence methods have been found immensely successful in applications in which big data are available. The applicability of these algorithms is being extensively investigated in scientific disciplines such as exploration geophysics with the goal of reducing exploration and development costs. In this context, we have used a special kind of unsupervised recurrent neural network and its variants, Hopfield neural networks and the Boltzmann machine, to solve the problems of Kirchhoff and reverse time migrations. We use the network to migrate seismic data in a least-squares sense using simulated annealing to globally optimize the cost function of the neural network. The weights and biases of the neural network are derived from the physics-based forward models that are used to generate seismic data. The optimal configuration of the neural network after training corresponds to the minimum energy of the network and thus gives the reflectivity solution of the migration problem. Using synthetic examples, we determine that (1) Hopfield neural networks are fast and efficient and (2) they provide reflectivity images with mitigated migration artifacts and improved spatial resolution. Specifically, the presented approach minimizes the artifacts that arise from limited aperture, low subsurface illumination, coarse sampling, and gaps in the data.

Resource Allocation Using SPA Based on Different Cost Functions in Elastic Optical Networks

Routing, modulation and spectrum allocation in elastic optical networks is a problem aiming at increasing the capacity of the network. Many algorithms such as shortest path algorithm can be used as the routing section of this problem. The efficiency of these algorithms is partly based on how the cost of each link is defined. In this study, we considered several basic metrics in cost of network links and compared their effects on the network capacity. In particular, the static costs and the dynamic costs were evaluated and compared. For dynamic scenarios, compared to static scenarios, at least one additional factor, the usage of the links, was added. We further considered a new factor that is based on probability of accommodating the signal at a given time in any given link. The results show that, among them, the shortest path algorithm provides the least blocking probability when the cost is a combination of link length and the abovementioned possibility/usage of the link.

Cooperative demand‐side management scenario for the low‐voltage network in liberalised electricity markets

This work presents a cooperative demand-side management (DSM) scenario in a low-voltage network considering a context of liberalised electricity markets. The authors show that introducing an additional inter-supplier cooperation mechanism among consumers enhances a better use of the flexibility consented by each individual, hence aiming at reaching a global optimum instead of optimising the costs of a few ones. To that end, a real-time pricing (RTP) scheme is explored based on cost functions differentiated in both time and consumption level that should reflect the true energy cost. The authors apply to each consumer a commodity cost function shared among the set of cooperative users of its respective supplier as well as one common network cost function shared by all cooperative users of all suppliers in the considered network. Each individual runs, through its smart meter (SM), a decentralised optimisation algorithm defining an energy consumption schedule for a set of flexible appliances (FAs). The mechanism the authors propose ensures that a fair cost distribution between all users is achieved by reaching the Nash equilibrium. To assess their proposition, they confront it to intermediate consumption strategies through a benchmark. The results confirm that their inter-supplier cooperation mechanism always leads to the minimum total cost.

Optimal infrastructure capacity of automated on-demand rail-bound transit systems

Fully-automated services potentially allow for greater flexibility in operations and lower marginal operational costs. The objective of this study is to determine the capacity requirements of an envisaged automated on-demand rail-bound transit system which offers a direct non-stop service. An optimization model is formulated for determining the optimal track and station platform capacities for an on-demand rail transit system so that passenger, infrastructure and operational costs are minimized. The macroscopic model allows for studying the underlying relations between technological, operational and demand parameters, optimal capacity settings and the obtained cost components. The model is applied to a series of numerical experiments followed by its application to part of the Dutch railway network. The performance is benchmarked against the existing service, suggesting that in-vehicle times can be reduced by 10% in the case study network while the optimal link and station capacity allocation is comparable to those currently available in the case study network. While network geometry and demand distribution are always the underlying determinants of both service frequencies and in-vehicle times, line configuration is only a determinant in the conventional system, whereas the automated on-demand rail service better caters for the prevailing demand relations, resulting in greater variations in service provision. A series of sensitivity analyses are performed to test the consequences of a range of network structures, technological capabilities, operational settings, cost functions and demand scenarios for future automated on-demand rail-bound systems.

Hybrid machine learning and optimisation method to solve a tri‐level road network protection problem

In this study, the authors employ machine learning to develop a new solution method for solving a tri-level network protection problem. In the upper-level, the planner aims to minimise the impact of the interdictor's attempt to disrupt a road network through protection activities. At the middle-level, however, the interdictor seeks to maximise the network's cost function, that is total travel time while the user equilibrium assignment models the road users behaviour at the lower-level. The proposed solution algorithm combines implicit enumeration with machine learning for faster performance. In so doing, four machine learning methods are evaluated among which the artificial neural network model shows the best performance and thereby to be exploited. Principal component analysis is also employed as part of the data pre-processing to perform dimensionality reduction. The proposed solution algorithm exhibits a reasonable level of tractability when employed to solve large problems in which a real-world network is under investigation. Although it cannot guarantee global optimality, it is argued that this is an essential compromise for the application of the network optimisation problems on extensive real-world networks and the large solution space that they generate.