Computational Experiments Research Articles

Construction of balanced sliced orthogonal arrays

Balanced sliced orthogonal arrays (BSOAs) are useful for designing sliced computer experiments with qualitative and quantitative factors or multiple models and nested computer experiments with two codes of different levels of accuracy. In this article, we propose a systematic method to construct BSOAs by making use of the A + B design structure and its modifications. A general method for construction of asymmetric BSOAs is also presented based on an orthogonal array with orthogonal partition. These methods are flexible for strength and factors. Numerous new symmetric and asymmetric BSOAs including a lot of ones with high strength can thus be constructed. Compared with the existing BSOAs, the constructed BSOAs have more number of columns. Moreover, these methods can also be used to construct BSOAs whose strength is different from the strength of their slices. Some selective BSOAs are tabulated for practical uses.

Computation of feedback control laws based on switched tracking of demonstrations

A common approach in robotics is to learn tasks by generalizing from special cases given by a so-called demonstrator. In this paper, we apply this paradigm and present an algorithm that uses a demonstrator (typically given by a trajectory optimizer) to automatically synthesize feedback controllers for steering a system described by ordinary differential equations into a goal set. The resulting feedback control law switches between the demonstrations that it uses as reference trajectories. In comparison to the direct use of trajectory optimization as a control law, for example, in the form of model predictive control, this allows for a much simpler and more efficient implementation of the controller. The synthesis algorithm comes with rigorous convergence and optimality results, and computational experiments confirm its efficiency.

Analysis of organic matter stock dynamics in arable soils of Yamal: simulation experiments with the ROMUL model

The goal of this work is to analyze the dynamics of soil organic matter (SOM) in arable soils of Yamal using computational experiments with the ROMUL model. Soil temperature and moisture dynamics were simulated using the SCLISS model. The Yamal experimental station was organized in 1932. The soil is Plaggic Podzol. The thickness of humus horizon (PY) is 30 cm. Pools, kg/m2: SOM – 9.32 and 13.75; N – 0.46 and 0.66 for 0–20 and 0–30 cm respectively. Background: soddy-green-moss tundra near Salekhard. Soil: Folic Podzol: litter (O, 2 cm) and humus-accumulative (AY, 4 cm) and illuvial ferruginous (BF, 6 cm) horizons. Pools, kg/m2: SOM – 1.38 and 2.69, N – 0.03 and 0.18 for 0–2 and 2–6 cm respectively. The quantity and quality of background fall was estimated according to literature data. Clarification of the amount of fall was carried out by the method of inverse problem solving (spin-up). Peat application was simulated: starting application in the dose of 12 kg/m2 and maintenance (every 6 years) in doses 4, 8 and 12 kg/m2, N 1%. Application of mineral fertilizers, 2 variants: a) N 4 g/m2 (40 kg/ha) in peatting years and b) application of the same dose of nitrogen every year. The duration of computational experiments is 30 and 90 years. Computational experiments showed that after the starting peatting of Folic Podzol at the rate of 12 kg/m2 after 30 years, only 15% of the applied detritus remained. Maintenance peatting of 8–12 kg/m2 leads to an increase in SOM pool to 20–30 kg/m2 and excessive accumulation of detritus. Application of 4 kg/m2 of peat once every 6 years for 90 years shows the dynamics of SOM pools from Folic Podzol to Plaggic Podzol which is confirmed by field survey data. Application of mineral nitrogen (4 g/m2) once every 6 years does not affect SOM pools. Annual application of mineral N (4 g/m2) increases SOM pools due to intensified humification. This fact requires experimental verification.

An Approach to a Priori Assessment of Fuzzy Classification Models in Monitoring Tasks

The article addresses the problems of using automation tools to perform monitoring and management tasks applicable to assessing the quality of fuzzy classification models, where the classification procedure is implemented on the basis of knowledge (rules) in the absence of the training set. An approach is proposed to obtain a priori assessments of the classification quality based on the study of the used model sensitivity to changes in the values of internal parameters during the corresponding modeling. The interpretation of the modeling results in the form of risk assessment caused by the self-imperfection of the classification models is obtained. The article provides an example of a fuzzy classification model based on a comparison of the current state of a monitoring object described using fuzzy features with a set of predefined typical states, which form corresponding fuzzy equal (close) states (monitoring situations). The comparison is carried out using the fuzzy implication operation provided that the required reliability is met. The example of this model demonstrates how the type of implication operation, as well as the internal features of the model, affect the results of classification, and appropriate indicators are proposed, which are both an interpretation of generally accepted indicators for assessing the classification quality, and unique, inherent in the considered model. Computational experiments were carried out, which made it possible to obtain graphs of changes in classification quality assessment indicators for the considered model and its modification and visualize the influence of internal parameters of the model on the results of its application. A number of indicators are proposed that allow for an a priori assessment of the risks arising from the application of the model before its actual application.

Cellulase exhibited a therapeutic potential to inhibit Salmonella enterica serovar Typhi biofilm by targeting multiple regulatory proteins of biofilm

Biofilm-mediated Salmonella enterica serovar Typhi (Salmonella Ser. Typhi) infections are a growing global health issue due to the formation of antibiotic resistance. The study aimed to discover some of the druggable target proteins of Salmonella Ser. Typhi biofilm and antibiofilm enzyme to prevent Salmonella Ser. Typhi biofilm-mediated infection. Enzymatic therapy has demonstrated effective therapeutic results against bacterial infections due to its specificity and high binding capacity to the target. Therefore, this study focused on the computational interaction between the cellulase enzyme and Salmonella Ser. Typhi biofilm targets proteins with help of the various computational experiments such as ADMET (absorption, distribution, metabolism, excretion, and toxicity), protein-protein interactions, MMGBSA, etc. Further, in vitro validations of the typhoidal biofilm and cellulose presence in Salmonella Ser. Typhi biofilm was conducted using Scanning Electron Microscopy (SEM), Fourier transform infrared spectroscopy, and Raman analysis. Additionally, a minimum biofilm inhibitory concentration assay for cellulase was conducted and find out the optimized cellulase concentration which showed its inhibitory effect on the Salmonella Ser. Typhi. The cellulase antibiofilm effect was analyzed with the help of SEM analysis. Further, the cellulose content in Salmonella Ser. Typhi was quantified before and after treatment of cellulase enzyme. As a result, 58.82 % cellulose content was decreased due to cellulase treatment in Salmonella Ser. Typhi. From the seven selected typhoidal biofilm regulatory proteins of Salmonella Ser. Typhi, we identified only five potential druggable targets: BcsA, CsgE, OmpR, CsgF, and CsgD. The BcsA protein is responsible for cellulose production in Salmonella Ser. Typhi biofilm. Consequently, cellulose worked as a fascinating drug target in Salmonella Ser. Typhi biofilm. Therefore, we used cellulase as a potential antibiofilm enzyme for target-based disruption of biofilm. The cellulase showed a high binding affinity with all five identified target proteins [BcsA(-205.62 kcal/mol) > CsgE(-108.20 kcal/mol) > OmpR(-107.58 kcal/mol) > CsgF(-73.74 kcal/mol) > CsgD(-66.61 kcal/mol)] in the protein-protein interaction analysis. Our computational analysis suggests that the cellulase enzyme may be used as a potential antibiofilm enzyme against Salmonella Ser. Typhi biofilm.

Размежевание или сплочение? (Динамика сетевой структуры политических телеграм-каналов: моделирование и эмпирический анализ)

In this paper, the authors propose novel methodology to analyze the dynamics of large-scale online network structures caused by significant exogenous shocks (foreign policy crises and the onset of military conflicts). The article focuses on diagnosing cohesion or polarization processes. As a first step, on the basis of the existing theoretical traditions, the authors analyze mechanisms at the micro-level that may lead to consolidation or polarization. Among these, they reveal mechanisms based on changes in position and mechanisms based on changes in identity. As a second step, the authors determine the “network projection” of these mechanisms’ actions: the changes in the structures of interaction networks between individuals they lead to. At this point mathematical modeling plays a key role, allowing for specific predictions regarding network behavior, expressed in changes in observable metrics. Four network indicators are proposed: the number of communities (clusters), modularity, the number of connections, and the average degree of nodes. As a result of computational experiments, the expected dynamics of all the indicators is calculated, which is compared with the corresponding mechanisms of changes in political positions or identity. As a third step, these predictions are compared with the indicators calculated from the empirical data. For the empirical analysis the article uses messages from the Russianlanguage political and news telegram channels, corresponding in time to the start of a special military operation (February-March, 2022), and those that chronologically occur 10 months earlier or later (April, 2021, and December, 2022). The general sample of such messages contains those with hyperlinks to other channels. It is the presence of such hyperlinks is considered to be a sign of the connection and is used to build a network. For each time period, the same network parameters are measured as for the computational data. Comparison of the observed dynamics with what is predicted by the models indicates that in the case of the special military operation, one key mechanism was at work — cohesion through the expansion of in-group boundaries, based on shifting to higher-order identity categories.

Strategic decision-making for sustainable production and distribution in automotive industry: a machine learning enabled dynamic multi-objective optimisation

ABSTRACT Over the last decade, numerous researchers have disclosed that major automotive companies do not conform to regulatory or societal expectations regarding their environmental and social performances. This paper explores the dynamic capabilities of production distribution within the sustainability practices of automotive industries. It offers insights to better grasp and articulate the environmental, economic, and social dimensions of sustainable supply chains. The research framework encloses all supply chain phases, from raw material sourcing to retailing finished products. Three conflicting objective functions are identified: social advantages maximisation, cost minimisation, and emission minimisation. Specifically, the study tackles a dynamic multi-objective optimisation model where each automobile type faces a series of dynamic demands. The dynamic nature of the problem poses significant challenges to conventional evolutionary algorithms for detecting the optimal solutions over time. Therefore, we introduce an interconnected prediction-based dynamic non-dominated sorting algorithm (ICP-DNSGA-II). Finally, extensive computational experiments are conducted to assess the effectiveness of this holistic approach. The findings offer valuable insights for automotive industry stakeholders and policymakers, illustrating its potential to enhance operational efficiency and sustainability performance across the supply chain. Most importantly, this paper proposes an automated decision-making approach to generate optimal solutions with dynamic changes in market demands.

Optimizing dedicated lanes and tolling schemes for connected and autonomous vehicles to address bottleneck congestion considering morning commuter departure choices

The introduction of connected and autonomous vehicles (CAVs) provides a significant opportunity to address the persistently increasing problem of urban traffic congestion. By virtue of their connectivity and automation features, CAVs can reduce vehicle headways, thereby increasing road capacity and enhancing throughput. It has been hypothesized that CAV-infrastructure design policies can influence traveler behavior in ways that could reduce congestion. This research focuses on the potential of using CAV-dedicated lanes (CAVL) to alleviate traffic congestion in a bottleneck corridor that serves both human-driven vehicles (HDVs) and CAVs. We delve into investigating the impacts of CAVLs on the departure time and lane choices of morning commuters. The study first expresses traffic equilibrium conditions as a linear program with complementarity constraints. Then, a system-optimal commute congestion management design is formulated to minimize the overall system cost, which consists of queuing delays and early and late arrival costs. The results of the computational experiments suggest that: (i) the CAV technological advancements can significantly reduce traffic congestion under CAVL deployment with an almost similar effect as a tolling policy; and (ii) the lower value of time for CAV commuters leads them to depart closer to their desired arrival time without a tolling policy, which could significantly increase the bottleneck traffic congestion that commuters experience, particularly HDVs.

Global stability analysis of an extended SUC epidemic mathematical model

Abstract In this study, we conduct a global stability analysis of an extended Susceptible-Unidentified infected-Confirmed (SUC) epidemic mathematical model. In the original SUC model, the entire population consists of individuals who are susceptible, those with unidentified infections, and those with confirmed infections, without accounting for births and deaths. In the proposed extended SUC model, we incorporate the dynamics of births and deaths into the original SUC model. We analyze the global stability of this extended SUC epidemic mathematical model and perform several computational experiments to validate the global stability analysis. Through this realistic extended SUC model, we aim to advance the current understanding of epidemiological modeling and provide valuable insights for guiding public health interventions and policies.

Bayesian Sequential Design of Computer Experiments for Quantile Set Inversion

We consider an unknown multivariate function representing a system—such as a complex numerical simulator—taking both deterministic and uncertain inputs. Our objective is to estimate the set of deterministic inputs leading to outputs whose probability (with respect to the distribution of the uncertain inputs) of belonging to a given set is less than a given threshold. This problem, which we call Quantile Set Inversion (QSI), occurs for instance in the context of robust (reliability-based) optimization problems, when looking for the set of solutions that satisfy the constraints with sufficiently large probability. To solve the QSI problem we propose a Bayesian strategy, based on Gaussian process modeling and the Stepwise Uncertainty Reduction (SUR) principle, to sequentially choose the points at which the function should be evaluated to efficiently approximate the set of interest. We illustrate the performance and interest of the proposed SUR strategy through several numerical experiments.

Bound-state confinement after trap-expansion dynamics in integrable systems

Abstract Integrable systems possess stable families of quasiparticles, which are composite objects (bound states) of elementary excitations. Motivated by recent quantum computer experiments, we investigate bound-state transport in the spin- 1 / 2 anisotropic Heisenberg chain (XXZ chain). Specifically, we consider the sudden vacuum expansion of a finite region A prepared in a non-equilibrium state. In the hydrodynamic regime, if interactions are strong enough, bound states remain confined in the initial region. Bound-state confinement persists until the density of unbound excitations remains finite in the bulk of A. Since region A is finite, at asymptotically long times bound states are ‘liberated’ after the ‘evaporation’ of the unbound excitations. Fingerprints of confinement are visible in the space-time profiles of local spin-projection operators. To be specific, here we focus on the expansion of the p-Néel states, which are obtained by repetition of a unit cell with p up spins followed by p down spins. Upon increasing p, the bound-state content is enhanced. In the limit p → ∞ one obtains the domain-wall initial state. We show that for p < 4, only bound states with n > p are confined at large chain anisotropy. For p ≳ 4 , bound states with n = p are also confined, consistent with the absence of transport in the limit p → ∞ . The scenario of bound-state confinement leads to a hierarchy of timescales at which bound states of different sizes are liberated, which is also reflected in the dynamics of the von Neumann entropy.

Modeling of Filtration and Heat and Mass Transfer Processes in Groundwater

Purpose. The method of mathematical modeling is an important tool for solving complex problems involving the analysis of groundwater dynamics and heat and mass transfer processes in them when studying their contamination from various anthropogenic sources in the event of accidental spills of chemically hazardous substances, etc. The main purpose of the article is to develop a set of mathematical models for calculating the process of filtration of non-pressure groundwater, mass transfer of impurities and the process of heat transfer in groundwater. Methodology. The two-dimensional Boussinesq equation of filtration was used to predict the dynamics of groundwater. The two-dimensional equation of convective-diffusive transport of impurities was used to model the processes of mass transfer in groundwater. The process of freezing of individual sections of the groundwater flow is modeled using the Laplace equation for the velocity potential (calculation of the flow velocity field in a time-varying geometry) and the two-dimensional equation of heat transfer in groundwater. Finite difference schemes were used to solve the modeling equations of groundwater dynamics and heat and mass transfer. Findings. A set of mathematical models has been developed to calculate the process of filtration of non-pressure groundwater and its chemical contamination. The experiment has confirmed the adequacy of the constructed numerical model of filtration of a non-pressure groundwater flow. An effective mathematical model was developed that allows determining the temperature fields in groundwater during the operation of a well used to freeze certain sections of the flow. The results of computer modeling indicate the effectiveness of the developed mathematical models. Originality. Effective mathematical models for predicting the level of chemical contamination of groundwater, its dynamics and thermal regime are proposed. The constructed mathematical models make it possible to determine the dynamics of changes in the temperature regime of groundwater during the operation of wells through which refrigerant is supplied to freeze individual areas. A computer program has been developed that allows for a comprehensive assessment of groundwater conditions. Practical value. A set of computer programs has been developed to conduct a computational experiment to study the processes of filtration, chemical contamination of groundwater and heat transfer processes in them. This set of programs can be used for the scientific substantiation of engineering solutions aimed at protecting groundwater.

Canonical cuts of path powers

The MaxCut problem aims to find a bipartition of vertices in a given graph such that the number of edges with one end vertex in each part is maximum among all bipartitions. NP-hardness when restricted to interval graphs has been recently announced. Surprisingly, all previously published attempts at polynomial-time algorithms for unit interval graphs turned out to be wrong, which justifies the search for subclasses where MaxCut can be handled. We introduce canonical cuts whose pattern allows an easy computation of the cut size for the power of paths $P_n^k$. Using canonical cuts, we calculate the structure and the size of maximum cuts for $k\leq 5$ and for $n\leq \frac{2}{3}(2k+2)$. We prove that the known size for a maximum cut for reduced co-bipartite chain graphs can be achieved by a canonical cut. We perform computational experiments on each $P_n^k$ graph with $1\leq k\leq n\leq 43$ and show that most of them allow a canonical cut that is maximum. We display a table with the found cases where there is no canonical cut which is a maximum cut. In these graphs, the difference between the maximum cut and some canonical is at most 3 units. This indicates canonical cuts as a good approach to tackle the maximum cut on $P_n^k$ graphs.

Optimization of Cost-Based Hybrid Flowshop Scheduling Using Teaching-Learning-Based Optimization Algorithm

A cost-based hybrid flowshop scheduling (CHFS) combines flow shop and job shop elements, with cost considerations as a key indicator. CHFS is a complex combinatorial optimization challenge encountered in real-world manufacturing and production environments. This paper investigates the optimization of a CHFS problem using the Teaching Learning-Based Optimization (TLBO) algorithm. Effective CHFS is crucial for achieving production balance, reducing costs, and improving customer satisfaction. The authors formulate the CHFS scheduling problem and propose applying the TLBO algorithm to minimize total costs, including labor, energy, maintenance, and delay expenses. The performance of the TLBO technique is evaluated through computational experiments on various CHFS problem instances. The results demonstrate the effectiveness of the TLBO algorithm, which achieved the best results in 42% of the test cases, surpassing other algorithms like the Grey Wolf Optimizer and Particle Swarm Optimization. Additionally, the TLBO algorithm had the highest average performance ranking across the comparative algorithms. The study highlights the potential of the TLBO algorithm as an efficient optimization tool for complex manufacturing scheduling problems.

Exact methods and a variable neighborhood search for the robust capacitated [formula omitted]-median problem

The capacitated p-median problem (CPMP) involves placing p identical facilities in a network and assigning customer nodes to these facilities to satisfy all customer demands with minimal transportation costs. In practical applications, demand and distance parameters are often uncertain during the planning process, leading to infeasible or excessively costly solutions if these uncertainties are disregarded. This paper addresses the robust CPMP (RCPMP), which incorporates demand uncertainty into the problem using the robust optimization paradigm. We propose a general framework to model and solve the RCPMP, considering different polyhedral uncertainty sets, namely the cardinality-constrained and the knapsack sets. We develop exact approaches encompassing compact models, different families of valid inequalities, and branch-and-cut and branch-and-price algorithms, exploring both implemented uncertainty sets and problem structure. Furthermore, we implement an efficient Variable Neighborhood Search (VNS) heuristic to solve these robust variants, which incorporates state-of-the-art algorithms, parallelization techniques, and optimized data structures. Computational experiments using adapted benchmark instances with up to 400 nodes indicate the effectiveness of the proposed approaches. The results show that using parallelization and hash tables within the VNS heuristic promotes significant performance improvements and yields near-optimal solutions for most instances, as well as outperforming the exact approaches in several instances where the optimal solution was not found. Moreover, these results highlight the benefits of using robust solutions in practical settings, especially when considering different uncertainty sets to generate solutions with advantageous trade-offs between cost and risk.

Predicting the Consequences of an Emergency at a Railway Station

Purpose. The paper considers the problem of determining the size of the damage zones in the event of an emergency at a railway station due to a tanker fire. The task of forecasting is to determine the zones of thermal pollution, as well as chemical and mechanical pollution. The main objective of the study is to create numerical models for calculating the zones of mechanical and thermal pollution in the event of a fire at a railway station. Methodology. To analyze the size and intensity of zones of thermal, chemical, and mechanical environmental pollution in the event of an extreme situation at a railway station, we used the equations of heat and mass transfer and Newton's second law for modeling mechanical environmental pollution. To solve the equations, numerical methods such as Euler's method and finite difference schemes were used. On the basis of the developed numerical models, a computer code was created to conduct a computational experiment. Findings. Modern computer models for assessing the zones of chemical, thermal, and mechanical pollution in the event of an extreme situation have been developed. The results of computer modeling are presented. Originality. A set of numerical models for computer simulation of heat and mass transfer processes and dynamics of point motion has been created, which allows conducting a computational experiment to determine the contamination zones during a fire at a railway station. Practical value. A computer code was developed on the basis of the created mathematical models. This code is a tool for solving important problems in the field of environmental safety and civil protection. The computer code makes it possible to quickly determine the intensity and size of environmental pollution zones in the event of an extreme situation.

PROBLEMS OF MANAGING THE IMMUNE RESPONSE UNDER CONDITIONS OF COMPETITIVE ADSORPTION, DIFFUSION PERTURBATIONS AND TEMPERATURE RESPONSE OF THE ORGANISM

Analysis of the problems associated with the rapid spread of infectious diseases indicates the need for new approaches to the diagnosis and develop- ment of effective personalized treatment programs. One of the important aspects of solving this problem is the creation of a mathematical modeling toolkit for predicting the dynamics of infectious diseases, taking into account spatial effects, concentrated effects, various mechanisms of body protec- tion in the conditions of the application of various types of therapeutic methods of treatment. An important component of complex methods of treat- ment of a wide range of infectious diseases is the use of adsorption drugs, which, along with detoxification of the body, are able to provide an addi- tional mechanism of neutralization of viral elements. In the process of using such means, not only pathogenic elements and toxins will be adsorbed, but also various types of immune agents, which will affect the dynamics of the immune response. In the paper the infectious disease model is generalized to take into account the features of competitive adsorption of antigens and immune agents in the conditions of diffusion perturbations, concentrated influences and temperature reaction of the body. By synthesizing the ideas of step-by-step procedures, asymptotic and numerical methods, an effective computational technology of step-by-step approximation of the solution of the original model singularly perturbed problem was built. The results of the computer experiments presented in the paper illustrate the features of reducing the predicted concentration of viral elements during their competi- tive adsorption together with immune agents, in particular, the effect of reducing the efficiency of the use of non-specific adsorbents. It is emphasized that taking into account the features of the action of competitive adsorption is important for making decisions regarding the formation of rational treatment programs with the complex use of adsorbing substances and immunological drugs.

Active Learning for a Recursive Non-Additive Emulator for Multi-Fidelity Computer Experiments

Computer simulations have become essential for analyzing complex systems, but high-fidelity simulations often come with significant computational costs. To tackle this challenge, multi-fidelity computer experiments have emerged as a promising approach that leverages both low-fidelity and high-fidelity simulations, enhancing both the accuracy and efficiency of the analysis. In this article, we introduce a new and flexible statistical model, the Recursive Non-Additive (RNA) emulator, that integrates the data from multi-fidelity computer experiments. Unlike conventional multi-fidelity emulation approaches that rely on an additive auto-regressive structure, the proposed RNA emulator recursively captures the relationships between multi-fidelity data using Gaussian process priors without making the additive assumption, allowing the model to accommodate more complex data patterns. Importantly, we derive the posterior predictive mean and variance of the emulator, which can be efficiently computed in a closed-form manner, leading to significant improvements in computational efficiency. Additionally, based on this emulator, we introduce four active learning strategies that optimize the balance between accuracy and simulation costs to guide the selection of the fidelity level and input locations for the next simulation run. We demonstrate the effectiveness of the proposed approach in a suite of synthetic examples and a real-world problem. An R package RNAmf for the proposed methodology is provided on CRAN.

Trainable Reference Spikes Improve Temporal Information Processing of SNNs With Supervised Learning.

Spiking neural networks (SNNs) are the next-generation neural networks composed of biologically plausible neurons that communicate through trains of spikes. By modifying the plastic parameters of SNNs, including weights and time delays, SNNs can be trained to perform various AI tasks, although in general not at the same level of performance as typical artificial neural networks (ANNs). One possible solution to improve the performance of SNNs is to consider plastic parameters other than just weights and time delays drawn from the inherent complexity of the neural system of the brain, which may help SNNs improve their information processing ability and achieve brainlike functions. Here, we propose reference spikes as a new type of plastic parameters in a supervised learning scheme in SNNs. A neuron receives reference spikes through synapses providing reference information independent of input to help during learning, whose number of spikes and timings are trainable by error backpropagation. Theoretically, reference spikes improve the temporal information processing of SNNs by modulating the integration of incoming spikes at a detailed level. Through comparative computational experiments, we demonstrate using supervised learning that reference spikes improve the memory capacity of SNNs to map input spike patterns to target output spike patterns and increase classification accuracy on the MNIST, Fashion-MNIST, and SHD data sets, where both input and target output are temporally encoded. Our results demonstrate that applying reference spikes improves the performance of SNNs by enhancing their temporal information processing ability.

Tuning parameters of deep neural network training algorithms pays off: a computational study

AbstractThe paper aims to investigate the impact of the optimization algorithms on the training of deep neural networks with an eye to the interaction between the optimizer and the generalization performance. In particular, we aim to analyze the behavior of state-of-the-art optimization algorithms in relationship to their hyperparameters setting to detect robustness with respect to the choice of a certain starting point in ending on different local solutions. We conduct extensive computational experiments using nine open-source optimization algorithms to train deep Convolutional Neural Network architectures on an image multi-class classification task. Precisely, we consider several architectures by changing the number of layers and neurons per layer, to evaluate the impact of different width and depth structures on the computational optimization performance. We show that the optimizers often return different local solutions and highlight the strong correlation between the quality of the solution found and the generalization capability of the trained network. We also discuss the role of hyperparameters tuning and show how a tuned hyperparameters setting can be re-used for the same task on different problems achieving better efficiency and generalization performance than a default setting.