Predefined Search Space Research Articles

WAOA: A hybrid whale-ant optimization algorithm for energy-efficient routing in wireless sensor networks

Wireless Sensor Networks (WSNs) are vital for collecting data from remote environments. Nevertheless, the limited energy resources of sensor nodes render energy-efficient routing a critical concern for the successful operation of WSNs. To address these concerns, clustering, and routing are essential tasks in WSNs; clustering aims to organize sensor nodes into groups or clusters to minimize energy usage and prolong the network's lifespan. On the other hand, routing involves determining the optimum paths for transmitting data from the source nodes to the destination nodes. Nonetheless, it has been established that the current energy-efficient routing problem is an NP-hard, requiring a trade-off between energy and overall network performance. In this paper, we proposed a Hybrid Whale-Ant Optimization Algorithm (WAOA) for energy-efficient routing in WSNs. The proposed WAOA utilizes the Whale Optimization Algorithm (WOA) to find the suitable cluster head in the predefined search space, while the Ant Colony Optimization (ACO) searches the optimal route from the source cluster sensors to the cluster head within its predefined space. Linear programming construction is employed to formulate optimization problems for cluster head selection and search for the optimal route. The performance analysis demonstrates that the proposed WAOA performs better than MOORP, MMABC, and AZEBR by 5.78 %,16.11 %, and 18.52 %, respectively, in terms of network lifetime.

Using machine learning techniques in inverse problems of acoustical oceanography

AbstractThe goal of the work presented here is to study a novel approach for inverting acoustic signals recorded in the marine environment for the estimation of environmental parameters of the water column and/or the seabed. The proposed approach is based on signal feature extraction using a discrete wavelet packet transform, applied to the measured signal, and hidden Markov models that exploit the sequential patterns of the signals. The signal feature is thereafter used in the framework of a mixture density network, which, after training with sets of simulated signals calculated within a predefined search space, provides conditional posterior distributions of the recoverable parameters. The technique is tested with two test cases corresponding to different types of inverse problems. The first case corresponds to a simple problem of geoacoustic inversion, while the second is referred to a, rather unusual, still interesting problem of recovering the shape of a seamount using long‐range acoustic data. Both test cases are based on simulated experiments. The inversion results obtained using the proposed scheme are compared with inversion results using statistical features of the acoustic signal, which is another inversion approach well documented in the literature and is also based on the wavelet packet transform of the measured signal.

Adaptive Bayesian contextual hyperband: A novel hyperparameter optimization approach

<p>Hyperparameter tuning plays a significant role when building a machine learning or a deep learning model. The tuning process aims to find the optimal hyperparameter setting for a model or algorithm from a pre-defined search space of the hyperparameters configurations. Several tuning algorithms have been proposed in recent years and there is scope for improvement in achieving a better exploration-exploitation tradeoff of the search space. In this paper, we present a novel hyperparameter tuning algorithm named adaptive Bayesian contextual hyperband (Adaptive BCHB) that incorporates a new sampling approach to identify best regions of the search space and exploit those configurations that produce minimum validation loss by dynamically updating the threshold in every iteration. The proposed algorithm is assessed using benchmark models and datasets on traditional machine learning tasks. The proposed Adaptive BCHB algorithm shows a significant improvement in terms of accuracy and computational time for different types of hyperparameters when compared with state-of-the-art tuning algorithms.</p>

Tree-shaped multiobjective evolutionary CNN for hyperspectral image classification

Convolutional neural networks (CNNs) have achieved significant performances in hyperspectral image (HSI) classification in recent years. However, designing a high-performance CNN depends on human expertise heavily, which usually takes considerable time and labor. With regard to reducing the burden of designing the networks, neural architecture search (NAS) has attracted increasing attention. A typical NAS approach aims to optimize the network architectures in a predefined search space with a suitable search algorithm automatically. However, the existing NAS work does not fully consider the spatial resolution and the spectral noise interference of HSIs. Furthermore, most NAS approaches use sequential blocks or cells to construct the networks, which are unsuitable for extracting multiscale features of HSIs and result in degraded performance. Considering the above challenges, we propose a tree-shaped multiobjective evolutionary CNN (TMOE-CNN) for HSI classification. An expanded search space is designed, which includes the image patch size and the channel number of the input image patches. A multibranch supernetwork structure is proposed, which resembles a tree as the fundamental architecture for the network block. The image patch size and the denoising strength of the input image patches can be established adaptively throughout the evolutionary search process. The tree-shaped networks can fuse multiscale features to enhance the capacity of the network for feature extraction. Additionally, we consider both the classification accuracy and the floating-point computational complexity in the environmental selection. It is helpful to find the networks with simple structure and low complexity while ensuring classification accuracy. Experiments on different HSI datasets show that TMOE-CNN can search CNNs with high accuracies and simple structures automatically.

Evolutionary Architecture Optimization for Retinal Vessel Segmentation.

Retinal vessel segmentation (RVS) is crucial in medical image analysis as it helps identify and monitor retinal diseases. Deep learning approaches have shown promising results for RVS, but designing optimal neural network architecture is challenging and time-consuming. Neural architecture search (NAS) is a recent technique that automates the design of neural network architectures within a predefined search space. This study proposes a new NAS method for U-shaped networks, MedUNAS, that discovers deep neural networks with high segmentation performance and lower inference time for RVS problem. We perform opposition-based differential evolution (ODE) and genetic algorithm (GA) to search for the best network structure and compare discrete and continuous encoding strategies on the proposed search space. To the best of our knowledge, this is the first NAS study that performs ODE for RVS problems. The results show that the MedUNAS ODE and GA yield the best and second-best results regarding segmentation performance with less than 50% of the parameters of U-shaped state-of-the-art methods on most of the compared datasets. In addition, the proposed methods outperform the baseline U-Net on four datasets with networks with up to 15 times fewer parameters. Furthermore, ablation studies are performed to evaluate the generalizability of the generated networks to medical image segmentation problems that differ from the trained domain, revealing that such networks can be effectively adapted to new tasks with fine-tuning. The MedUNAS can be a valuable tool for automated and efficient RVS in clinical practice.

NPENAS: Neural Predictor Guided Evolution for Neural Architecture Search.

Neural architecture search (NAS) adopts a search strategy to explore the predefined search space to find superior architecture with the minimum searching costs. Bayesian optimization (BO) and evolutionary algorithms (EA) are two commonly used search strategies, but they suffer from being computationally expensive, challenging to implement, and exhibiting inefficient exploration ability. In this article, we propose a neural predictor guided EA to enhance the exploration ability of EA for NAS (NPENAS) and design two kinds of neural predictors. The first predictor is a BO acquisition function for which we design a graph-based uncertainty estimation network as the surrogate model. The second predictor is a graph-based neural network that directly predicts the performance of the input neural architecture. The NPENAS using the two neural predictors are denoted as NPENAS-BO and NPENAS-NP, respectively. In addition, we introduce a new random architecture sampling method to overcome the drawbacks of the existing sampling method. Experimental results on five NAS search spaces indicate that NPENAS-BO and NPENAS-NP outperform most existing NAS algorithms, with NPENAS-NP achieving state-of-the-art performance on four of the five search spaces.

The joint application of a metaheuristic algorithm and a Bayesian statistics approach for uncertainty and stability assessment of nonlinear magnetotelluric data

Abstract. In this paper, we have developed three algorithms, namely hybrid weighted particle swarm optimization (wPSO) with the gravitational search algorithm (GSA), known as wPSOGSA; GSA; and PSO in MATLAB to interpret one-dimensional magnetotelluric (MT) data for some corrupted and non-corrupted synthetic data, as well as two examples of MT field data over different geological terrains: (i) geothermally rich area, island of Milos, Greece, and (ii) southern Scotland due to the occurrence of a significantly high electrical conductivity anomaly under crust and upper mantle, extending from the Midland Valley across the Southern Uplands into northern England. Even though the fact that many models provide a good fit in a large predefined search space, specific models do not fit well. As a result, we used a Bayesian statistical technique to construct and assess the posterior probability density function (PDF) rather than picking the global model based on the lowest misfit error. The study proceeds using a 68.27 % confidence interval for selecting a region where the PDF is more prevalent to estimate the mean model which is more accurate and close to the true model. For illustration, correlation matrices show a significant relationship among layer parameters. The findings indicate that wPSOGSA is less sensitive to model parameters and produces more stable and reliable results with the least uncertainty in the model, compatible with existing borehole samples. Furthermore, the present methods resolve two additional geologically significant layers, one highly conductive (less than 1.0 Ωm) and another resistive (300.0 Ωm), over the island of Milos, Greece, characterized by alluvium and volcanic deposits, respectively, as corroborated by borehole stratigraphy.

Restricted Search Space Exploration With Refinement for Symbol Detection in Uplink Massive MIMO

Massive multiple-input multiple-output (MIMO) is a promising technique to overcome the explosive increase in the demand for high-speed data, quality of service and energy efficiency for fifth generation (5 G) and beyond wireless systems. Low-complexity symbol detection in massive MIMO is one of the challenging issues for their practical realization. The existing approximate matrix inversion based and matrix inversion less iterative symbol detection techniques suffer significant performance loss when the number of users increases in the system. Therefore, this article proposes a low-complexity restricted search space exploration based ordered iterative detection (REOD) technique to achieve superior performance in large user massive MIMO systems. The proposed low-complexity REOD technique iteratively prunes the discrepancy term associated with the previous estimation and further explores an improved estimate confined to a predefined search space. Convergence of the proposed REOD technique is analytically justified. An analytical expression for the approximate upper bound on the bit error rate is derived and corroborated by simulations. The obtained results prove the viability of the proposed REOD technique compared to the several existing state-of-the-art uplink massive MIMO detection techniques, both in terms of error performance and computational complexity.

Rapid Guessing in Low-Stakes Assessments: Finding the Optimal Response Time Threshold with Random Search and Genetic Algorithm

Rapid guessing is an aberrant response behavior that commonly occurs in low-stakes assessments with little to no formal consequences for students. Recently, the availability of response time (RT) information in computer-based assessments has motivated researchers to develop various methods to detect rapidly guessed responses systematically. These methods often require researchers to identify an RT threshold subjectively for each item that could distinguish rapid guessing behavior from solution behavior. In this study, we propose a data-driven approach based on random search and genetic algorithm to search for the optimal RT threshold within a predefined search space. We used response data from a low-stakes math assessment administered to over 5000 students in 658 schools across the United States. As we demonstrated how to use our data-driven approach, we also compared its performance with those of the existing threshold-setting methods. The results show that the proposed method could produce viable RT thresholds for detecting rapid guessing in low-stakes assessments. Moreover, compared with the other threshold-setting methods, the proposed method yielded more liberal RT thresholds, flagging a larger number of responses. Implications for practice and directions for future research were discussed.

Automated Anomaly Detection via Curiosity-Guided Search and Self-Imitation Learning.

Anomaly detection is an important data mining task with numerous applications, such as intrusion detection, credit card fraud detection, and video surveillance. However, given a specific complicated task with complicated data, the process of building an effective deep learning-based system for anomaly detection still highly relies on human expertise and laboring trials. Also, while neural architecture search (NAS) has shown its promise in discovering effective deep architectures in various domains, such as image classification, object detection, and semantic segmentation, contemporary NAS methods are not suitable for anomaly detection due to the lack of intrinsic search space, unstable search process, and low sample efficiency. To bridge the gap, in this article, we propose AutoAD, an automated anomaly detection framework, which aims to search for an optimal neural network model within a predefined search space. Specifically, we first design a curiosity-guided search strategy to overcome the curse of local optimality. A controller, which acts as a search agent, is encouraged to take actions to maximize the information gain about the controller's internal belief. We further introduce an experience replay mechanism based on self-imitation learning to improve the sample efficiency. Experimental results on various real-world benchmark datasets demonstrate that the deep model identified by AutoAD achieves the best performance, comparing with existing handcrafted models and traditional search methods.

SSSL: Secure Search Space Locking of Behavioral IPs

Raising the level of VLSI design abstraction from the Register Transfer Level (RTL) to the behavioral level has multiple advantages: (i) It reduces the turn-around-time, (ii) allows faster verification and (iii) extends the re-usability of the design as High-Level Synthesis (HLS) automatically re-optimizes the synthesized circuit when new process technologies are available by simply selecting a different technology library. Moreover, HLS makes extensive use of synthesis directives that control how to synthesize mainly loops (unroll or pipeline), arrays (register or RAM) and functions (inline or not). This further increases the re-usability of the behavioral code as it enables the generation of micro-architectures with different area vs. performance trade-offs. These advantages open the door to third party IP (3PIP) vendors providing Behavioral IPs (BIPs). Unfortunately, the market of third party BIPs is still very small and mostly limited to the HLS vendors themselves. Being so flexible is also their main weakness as it makes them only economically viable if the BIP provider can charge a large premium as it is highly unlikely that the BIP consumer will require their service again. Traditional IP vendors discriminate the price of the IP based on the amount of flexibility of the IP, e.g., RTL description vs. providing a synthesized gate netlist. We envision a similar price discrimination strategy for BIPs by limiting the re-usability of the BIP by partially encrypting the BIP source code. The main idea is to limit the search space, and hence, the re-usability of the BIP such that it only allows the BIP consumer to generate micro-architectures within a pre-defined search space range. This is accomplished by selectively fixing some of the synthesis directives in the form of pragmas at the source code while leaving others explorable. By encrypting the portion of the BIP that contains the fix pragmas we can guarantee that no designs outside of the pre-defined search space are generated. We believe that this work could serve as catalyst to grow the BIP market.

SiamBAN: Target-Aware Tracking With Siamese Box Adaptive Network.

Variation of scales or aspect ratios has been one of the main challenges for tracking. To overcome this challenge, most existing methods adopt either multi-scale search or anchor-based schemes, which use a predefined search space in a handcrafted way and therefore limit their performance in complicated scenes. To address this problem, recent anchor-free based trackers have been proposed without using prior scale or anchor information. However, an inconsistency problem between classification and regression degrades the tracking performance. To address the above issues, we propose a simple yet effective tracker (named Siamese Box Adaptive Network, SiamBAN) to learn a target-aware scale handling schema in a data-driven manner. Our basic idea is to predict the target boxes in a per-pixel fashion through a fully convolutional network, which is anchor-free. Specifically, SiamBAN divides the tracking problem into classification and regression tasks, which directly predict objectiveness and regress bounding boxes, respectively. A no-prior box design is proposed to avoid tuning hyper-parameters related to candidate boxes, which makes SiamBAN more flexible. SiamBAN further uses a target-aware branch to address the inconsistency problem. Experiments on benchmarks including VOT2018, VOT2019, OTB100, UAV123, LaSOT and TrackingNet show that SiamBAN achieves promising performance and runs at 35 FPS.

Design & Development of MPPT Using PSO With Predefined Search Space Based on Fuzzy Fokker Planck Solution

Production of clean, green solar PV (SPV) power in developing countries now becomes a trend because of their economic and technical benefits. Therefore, generating maximum power out of the SPV is a key searchable area. The SPV must produce power at its terminal at their maximum possible power. To reach to the maximum possible power, maximum power point tracker is used in conjunction with SPV. Extracting maximum power from SPV under varying partial shading condition is one of the important factor in performance improvement of SPV. The characteristics of classical MPPT controller is not acceptable under variable shading condition. A clear distinction between global maxima power point from global minima using MPPT technique must be needed for extracting maximum power. This paper proposes a PO MPPT based particle swarm optimization with improved search space, optimised through Fuzzy Fokker Planck solution. The pre-defined search space has been introduced to provide fine tune to membership function used in Fuzzy logic controller. The partial shading performance has been examined under four different condition such as active partial shading, colour spectrum, dust level and GHG concentration. Both hardware and simulation studies has been carried out for the proposed techniques. The MATLAB simulation result and that of proposed MPPT, offer more and better performance in terms of algorithm convergence by enhancing the efficiency of system under varying shading condition.

Automated search space and search strategy selection for AutoML

Existing works on Automated Machine Learning (AutoML) are mainly based on predefined search space. This paper seeks synergetic automation of two ingredients, i.e., search space and search strategies. Specifically, we formulate the automation of search space and search strategies as a combinatorial optimization problem. Our empirical study on many architecture benchmarks shows that identifying the suitable search space exerts more effect than choosing a sophisticated search strategy. Motivated by this, we attempt to leverage a machine learning method to solve the discrete optimization problem, and thus develop a Layered Architecture Search Tree (LArST) approach to synergize these two components. In addition, we use a probe model-based method to extract dataset-wise features, i.e., meta-features, which is able to facilitate the estimation of proper search space and search strategy for a given task. Experimental results show the efficacy of our approach under different search mechanisms and various datasets and hardware platforms.

Particle swarm optimization of partitions and fuzzy order for fuzzy time series forecasting of COVID-19

Major hyperparameters which affect fuzzy time series (FTS) forecasting are the number of partitions, length of partition intervals in the universe of discourse, and the fuzzy order. There are very few studies which have considered an integrated solution to optimize all the hyperparameters. In this paper, we strive to achieve optimum values of all three hyperparameters for fuzzy time series forecasting of the COVID-19 pandemic using the Particle Swarm Optimization (PSO) algorithm. We specifically propose two techniques, namely nested FTS-PSO and exhaustive search FTS-PSO for determining the optimal interval length, as an augmentation to the FTS-PSO model that optimizes the interval length and the fuzzy order. Nested PSO has two PSO loops: (i) the inner PSO optimizes the combination of fuzzy order and boundaries of intervals for a given number of partitions defined by the outer loop, and the resultant cost is fed back to the outer PSO; (ii) the outer PSO optimizes the number of partitions to reduce the cost while meeting the defined constraint. Exhaustive search FTS-PSO also has two loops where the inner loop is similar to nested FTS-PSO while the outer loop iterates over a pre-defined search space of number of partitions. We analyze the effectiveness of the two approaches by comparing with ARIMA, FbProphet, and the state-of-the-art FTS and FTS-PSO models. We adopt COVID-19 highly affected 10 countries worldwide to perform forecasting of coronavirus confirmed cases. We consider two phases of COVID-19 spread, one from the year 2020 and another from 2021. Our study provides an analytical aspect of the COVID-19 pandemic, and aims to achieve optimal number and length of intervals along with fuzzy order for FTS forecasting of COVID-19. The results prove that the exhaustive search FTS-PSO outperformed all the methods whereas nested FTS-PSO performed moderately well.

Exploring Neural Architecture Search Space via Deep Deterministic Sampling

Recent developments in Neural Architecture Search (NAS) resort to training the supernet of a predefined search space with weight sharing to speed up architecture evaluation. These include random search schemes, as well as various schemes based on optimization or reinforcement learning, in particular policy gradient, that aim to optimize a parametric architecture distribution and the shared model weights simultaneously. In this paper, we focus on efficiently exploring the important region of a neural architecture search space with reinforcement learning. We propose Deep Deterministic Architecture Sampling (DDAS) based on deep deterministic policy gradient and the actor-critic framework, to selectively sample important architectures in the supernet for training. Through balancing exploitation and exploration, DDAS is designed to combat the disadvantages of prior random supernet warm-up schemes and optimization schemes. Gradient-based NAS approaches require the execution of multiple short experiments in order to combat the random stochastic nature of gradient descent, while still only producing a single architecture. Contrary to this approach, DDAS employs a reinforcement learning-based agent and focuses on discovering a Pareto frontier containing many architectures over the course of a single experiment requiring 1 GPU day. Experimental results for CIFAR-10 and CIFAR-100 on the DARTS search space show that DDAS can depict in a single search, the accuracy-FLOPs (or model size) Pareto frontier, which outperforms random sampling and search. With a test accuracy of 97.27%, the best architecture found on CIFAR-10 outperforms the original second-order DARTS while using 600M fewer parameters. Additionally, DDAS finds an architecture capable of achieving 82.00% test accuracy on CIFAR-100 while using only 3.14M parameters and outperforming GDAS.

AutoRSISC: Automatic design of neural architecture for remote sensing image scene classification

• A remote sensing image scene classification algorithm base on neural architecture search (NAS) is proposed. • A search strategy is proposed to search for the most appropriate neural architecture in the pre-defined search space. • A performance estimation strategy to evaluate the performance of search strategy. • The proposed method saves time and manpower while achieves better accuracy compared with the manually design method. Remote sensing image scene classification (RSISC) is one of the bases of visual question answering for remote sensing data (RSVQA). In the conventional deep learning based solutions of RSISC, domain experts have to manually design the neural architecture, which is costly in both time and manpower. In this paper, AutoRSISC, a high resolution remote sensing image scene classification method based on neural architecture search (NAS) is proposed. AutoRSISC algorithm samples the neural architecture in a certain proportion to reduce the redundancy in the search space which is based on the continuous relaxation of the neural architecture representation. Edge normalization is introduced to make architecture search more efficient. The conventional deep learning approaches take days or more just to manually design the appropriate neural architecture according to the current data set. The experimental results show that our method saves time and manpower compared with the existing manually designed RSISC methods. Our AutoRSISC algorithm took 7.6 h from the automatic design of the most appropriate neural architecture to the end of all experiments according to the UCM data set, and the classification accuracy rate of the final test set reached 97.85%. And our AutoRSISC algorithm took 101.7 h from the automatic design of the most appropriate neural architecture to the end of all experiments according to the NWPU45 data set, and the classification accuracy rate of the final test set reached 94.66%.

MODULAR SEARCH SPACE FOR AUTOMATED DESIGN OF NEURAL ARCHITECTURE

The past years of research have shown that automated machine learning and neural architecture search are an inevitable future for image recognition tasks. In addition, a crucial aspect of any automated search is the predefined search space. As many studies have demonstrated, the modularization technique may simplify the underlying search space by fostering successful blocks’ reuse. In this regard, the presented research aims to investigate the use of modularization in automated machine learning. In this paper, we propose and examine a modularized space based on the substantial limitation to seeded building blocks for neural architecture search. To make a search space viable, we presented all modules of the space as multisectoral networks. Therefore, each architecture within the search space could be unequivocally described by a vector. In our case, a module was a predetermined number of parameterized layers with information about their relationships. We applied the proposed modular search space to a genetic algorithm and evaluated it on the CIFAR-10 and CIFAR-100 datasets based on modules from the NAS-Bench-201 benchmark. To address the complexity of the search space, we randomly sampled twenty-five modules and included them in the database. Overall, our approach retrieved competitive architectures in averaged 8 GPU hours. The final model achieved the validation accuracy of 89.1% and 73.2% on the CIFAR-10 and CIFAR- 100 datasets, respectively. The learning process required slightly fewer GPU hours compared to other approaches, and the resulting network contained fewer parameters to signal lightness of the model. Such an outcome may indicate the considerable potential of sophisticated ranking approaches. The conducted experiments also revealed that a straightforward and transparent search space could address the challenging task of neural architecture search. Further research should be undertaken to explore how the predefined knowledge base of modules could benefit modular search space.

Dimensionality reduction and identification of valid parameter bounds for the efficient calibration of automated driving functions

The industrialization of automated driving functions according to level 3 requires an efficient test and calibration concept to deal with an increased complexity, growing customer demands, and a larger vehicle fleet offered. Therefore, a method for a complexity reduction of the calibration parameter space is presented. In the two-step approach, a qualitative sensitivity analysis is used to identify valid regions in the search space and subsequently decrease dimensionality based on the parameter-specific global influences. The reduced parameter space and sensitivity information can then serve as a starting point for an efficient calibration process on the target hardware. To examine the method’s potential, our approach is applied to the parameter space of an automated driving function. The results expose clear dependencies between parameters and driving scenarios and allow an exclusion of parameter space dimensions based on sensitivity values. The predefined search space can be narrowed down to valid regions using the parameter range identification approach. Finally, the findings are validated with a quantitative variance-based sensitivity analysis. The validation confirms that our method provides equivalent results with a comparably smaller number of system evaluations.

Evolutionary Approach to Constructing a Deep Feedforward Neural Network for Prediction of Electronic Coupling Elements in Molecular Materials.

We present a general framework for the construction of a deep feedforward neural network (FFNN) to predict distance and orientation dependent electronic coupling elements in disordered molecular materials. An evolutionary algorithm automatizes the selection of an optimal architecture of the artificial neural network within a predefined search space. Systematic guidance, beyond minimizing the model error with stochastic gradient descent based backpropagation, is provided by simultaneous maximization of a model fitness that takes into account additional physical properties, such as the field-dependent carrier mobility. As a prototypical system, we consider hole transport in amorphous tris(8-hydroxyquinolinato)aluminum. Reference data for training and validation is obtained from multiscale ab initio simulations, in which coupling elements are evaluated using density-functional theory, for a system containing 4096 molecules. The Coulomb matrix representation is chosen to encode the explicit molecular pair coordinates into a rotation and translation invariant feature set for the FFNN. The final optimized deep feedforward neural network is tested for transport models without and with energetic disorder. It predicts electronic coupling elements and mobilities in excellent agreement with the reference data. Such a FFNN is readily applicable to much larger systems at negligible computational cost, providing a powerful surrogate model to overcome the size limitations of the ab initio approach.