Regression Scheme Research Articles

Wave episode based Gaussian process regression for extreme event statistics in ship dynamics: Between the Scylla of Karhunen–Loève convergence and the Charybdis of transient features

For a number of structural and fatigue problems in marine science, engineers need accurate statistics for kinematic and dynamical quantities, such as displacements and bending moments. For many such problems the response statistics depend nonlinearly on input stochastic processes, such as sea surface elevation. Nonlinearity prevents the use of standard linear methods, leading to the adoption of expensive experiments and time-consuming numerical simulations. In order to avoid this cost we present a machine learning framework to minimize the training set requirements. This framework consists of two parts. First, we use the Karhunen–Loève theorem to represent stochastic sea states with finite-time wave episodes, which have low dimensionality that nonetheless captures the features important to hydrodynamics and structural mechanics. However, the choice of the wave episode duration is ‘caught’ between the Scylla of slow Karhunen–Loève series convergence for long time durations and the Charybdis of missing transient behaviors when the interval is short. To combat this dilemma, we propose a division into a region, designed for parametric interpolation and machine learning, and a stochastic prelude region, designed to probabilistically model transients. The second part of the framework consists of a Gaussian Process Regression (GPR) model designed to learn the mapping from wave episodes to structural outputs. GPR is able to take advantage of the low dimensional parametric representation of the sea state in order to converge with reasonably-sized training sets (on the order n≈100). At the same time, we use a low-dimensional representation in order to represent the stochastic response time series. The principal advantages of the Gaussian process surrogate are the blazing speed of evaluation – ten thousand times faster than the direct method – and the built in uncertainty quantification. Taken together, we can reconstruct the statistics of the responses by sampling sea states via the Karhunen–Loève construction, estimating the corresponding model outputs using the trained GPR, and estimating statistics of interest through Monte-Carlo calculation on the surrogate model. Finally the developed framework allows for trivial computation of the statistics for different sea spectra, i.e. without the need to retrain the Gaussian Process Regression scheme.

A Transformer-Based Regression Scheme for Forecasting Significant Wave Heights in Oceans

In this article, we present a novel approach for forecasting significant wave heights in oceanic waters. We propose an algorithm based on the WaveWatch III, differencing, and a transformer neural network (Transformer). The data becomes stationary after first-order differencing, performed with the observed significant wave height and the wave height forecasts obtained from WaveWatch III. We perform a case study on a group of 92 buoys using WaveWatch III hindcasts. The Transformer model then provides the statistical forecasts of the residuals. The Transformer-based proposed framework obtains the root mean square error of 0.231 m for two days ahead forecasting. Our proposed method outperforms existing state-of-the-art machine learning and numerical approaches for significant wave heights prediction. Our results suggest that combining numerical and machine learning approaches gives better performance than using either alone.

Quantitative analysis of metal scraps using laser-induced breakdown spectroscopy measurements via unsupervised adversarial domain adaptation

In this study, we propose a domain adaptation-based scheme for the compositional analysis of metal scraps using laser-induced breakdown spectroscopy (LIBS) measurements. LIBS analysis along with machine learning models has demonstrated feasibility for estimating elemental compositions of various materials such as metal scraps. However, LIBS-captured spectra can significantly vary depending on the physical and chemical properties of samples, configuration of the experimental setup, and prevailing environmental conditions such as temperature or humidity levels. Consequently, schemes based on calibration methods and the formulation of experimental parameters have been previously proposed by assuming that the training set is acquired under the same setup and conditions as the test set. Furthermore, current transfer learning solutions simply limit the pretrained model on the new test set, thus requiring a labeled test set in advance for finetuning. To address these limitations, this study proposes a novel transfer learning-based approach that depends on the domain adaptation of a source domain to transfer knowledge and reduce the difference between domains, and thus improving the performance of the target learner. We used LIBS measurements from certified reference materials (CRMs) to pretrain a source model to estimate the elemental concentration with known compositions of CRM metal samples. Then, the pretrained model is adapted in an unsupervised manner for the test/target dataset, and the adapted target model is used to quantify the elemental concentration of metal scraps. For evaluation, the CRMs of five representative metal types, including aluminum, copper, iron, stainless steel, and brass of known compositions, are measured under a typical LIBS setup to develop the pretrained model. For unsupervised domain adaptation and testing, metal scraps from actual industrial fields are tested using LIBS under different conditions such as lasers of different wavelengths and moving stages. The experimental results demonstrate that the proposed scheme achieves a root mean square error of 1.47 wt% or less for all test datasets, which is significantly better than the results of conventional regression schemes. The proposed quantitative analysis approach can be generalized to the regression of metal scraps and other industrial applications using LIBS.

Computational assessment of groundwater salinity distribution within coastal multi-aquifers of Bangladesh

The rising salinity trend in the country’s coastal groundwater has reached an alarming rate due to unplanned use of groundwater in agriculture and seawater seeping into the underground due to sea-level rise caused by global warming. Therefore, assessing salinity is crucial for the status of safe groundwater in coastal aquifers. In this research, a rigorous hybrid neurocomputing approach comprised of an Adaptive Neuro-Fuzzy Inference System (ANFIS) hybridized with a new meta-heuristic optimization algorithm, namely Aquila optimization (AO) and the Boruta-Random forest feature selection (FS) was developed for estimating the salinity of multi-aquifers in coastal regions of Bangladesh. In this regard, 539 data samples, including ten water quality indices, were collected to provide the predictive model. Moreover, the individual ANFIS, Slime Mould Algorithm (SMA), and Ant Colony Optimization for Continuous Domains (ACOR) coupled with ANFIS (i.e., ANFIS-SMA and ANFIS-ACOR) and LASSO regression (Lasso-Reg) schemes were examined to compare with the primary model. Several goodness-of-fit indices, such as correlation coefficient (R), the root mean squared error (RMSE), and Kling-Gupta efficiency (KGE) were used to validate the robustness of the predictive models. Here, the Boruta-Random Forest (B-RF), as a new robust tree-based FS, was adopted to identify the most significant candidate inputs and effective input combinations to reduce the computational cost and time of the modeling. The outcomes of four selected input combinations ascertained that the ANFIS-OA regarding the best accuracy in terms of (R = 0.9450, RMSE = 1.1253 ppm, and KGE = 0.9146) outperformed the ANFIS-SMA (R = 0.9406, RMSE = 1.1534 ppm, and KGE = 0.8793), ANFIS-ACOR (R = 0.9402, RMSE = 1.1388 ppm, and KGE = 0.8653), Lasso-Reg (R = 0.9358), and ANFIS (R = 0.9306) models. Besides, the first candidate input combination (C1) by three inputs, including Cl− (mg/l), Mg2+ (mg/l), Na+ (mg/l), yielded the best accuracy among all alternatives, implying the role importance of (B-RF) feature selection. Finally, the spatial salinity distribution assessment in the study area ascertained the high predictability potential of the ANFIS-OA hybrid with B-RF feature selection compared to other paradigms. The most important novelty of this research is using a robust framework comprised of the non-linear data filtering technique and a new hybrid neuro-computing approach, which can be considered as a reliable tool to assess water salinity in coastal aquifers.

A real-time collaborative machine learning based weather forecasting system with multiple predictor locations

Weather forecasting is an important application in meteorology and has been one of the most scientifically and technologically challenging problems around the world. As the drastic effects of climate change continue to unfold, localised short term weather prediction with high accuracy has become more important than ever. In this paper, a collaborative machine learning-based real-time weather forecasting system has been proposed whereby data from several locations are used to predict the weather for a specific location. In this work, five machine learning algorithms have been used and tests have been performed in four different locations in Mauritius to predict weather parameters such as Temperature, Wind Speed, Wind Direction, Pressure, Humidity, and Cloudiness. The weather data were collected using the OpenWeather API from a mobile as well as a desktop edge device. The data were stored as a JSON file in both the IBM Cloudant database and a local MySQL database. Analytics were performed on both a local server that captures the incoming data from the edge device and via a servlet deployed on the IBM cloud platform. Five machine learning algorithms namely Multiple Linear Regression (MLP), Multiple Polynomial Regression (MPR), K-Nearest Neighbours (KNN), Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN) were tested using both collaborative and non-collaborative methods. The experiments showed that the collaborative regression schemes achieved 5% lower Mean Absolute Percentage Error (MAPE) than non-collaborative ones and the Multiple Polynomial Regression (MLR) algorithm outperformed all the other algorithms with errors ranging from 0.009% to 9% for the different weather parameters. In general, the results showed that collaborative based weather forecasting with multiple predictor locations can potentially increase the accuracy of the predictions in machine learning algorithms.

ACCEL: an efficient and privacy-preserving federated logistic regression scheme over vertically partitioned data

ACCEL: an efficient and privacy-preserving federated logistic regression scheme over vertically partitioned data

Estimation of Knee Extension Force Using Mechanomyography Signals Based on GRA and ICS-SVR

During lower-extremity rehabilitation training, muscle activity status needs to be monitored in real time to adjust the assisted force appropriately, but it is a challenging task to obtain muscle force noninvasively. Mechanomyography (MMG) signals offer unparalleled advantages over sEMG, reflecting the intention of human movement while being noninvasive. Therefore, in this paper, based on MMG, a combined scheme of gray relational analysis (GRA) and support vector regression optimized by an improved cuckoo search algorithm (ICS-SVR) is proposed to estimate the knee joint extension force. Firstly, the features reflecting muscle activity comprehensively, such as time-domain features, frequency-domain features, time–frequency-domain features, and nonlinear dynamics features, were extracted from MMG signals, and the relational degree was calculated using the GRA method to obtain the correlation features with high relatedness to the knee joint extension force sequence. Then, a combination of correlated features with high relational degree was input into the designed ICS-SVR model for muscle force estimation. The experimental results show that the evaluation indices of the knee joint extension force estimation obtained by the combined scheme of GRA and ICS-SVR were superior to other regression models and could estimate the muscle force with higher estimation accuracy. It is further demonstrated that the proposed scheme can meet the need of muscle force estimation required for rehabilitation devices, powered prostheses, etc.

Assessment of thermal conductivity of polyethylene glycol-carbon dot nanofluid through a combined experimental-data mining investigation

The aim of this study is to determine the effect of nanoparticle concentration (φ) and temperature on the thermal conductivity of polyethylene glycol (PEG)-carbon dot nanofluid (NF). The considered range for temperature and φ is 20–60 °C and 0–7%, respectively. The results indicated an ascending trend of NF thermal conductivity with boosting both temperature and φ. The percentage of increase in thermal conductivity of NF with temperature and φ compared to the base fluid was in the range of 7.23–13.43% and 75.08–85.17%, respectively. Moreover, two efficient data-driven approaches, namely Multi-variate linear regression (MLR) and response surface methodology (RSM) schemes were developed to simulate the thermal conductivity of the PEG-carbon dot NF and fit the predictive relationships. The outcomes of modeling revealed that RSM model (R = 0.984 and RMSE = 0.013 W/m.K), due to taking into account the interaction between volume fraction and temperature, yielded more accuracy than MLR (R = 0.960 and RMSE = 0.021 W/m.K).

Ultrasound estimation of strain time constant and vascular permeability in tumors using a CEEMDAN and linear regression-based method

Ultrasound poroelastography focuses on the estimation of the spatio-temporal mechanical behavior of tissues using data often corrupted with non-stationary noise. The cumulative strain calculated from prolonged temporal acquisition of RF data can face the problem of aggregate noise. This noise can significantly affect the accuracy of curve fitting techniques necessary to estimate the clinically significant strain Time Constant (TC) and related parameters. We present a new technique, which decomposes the non-linear temporal behavior of the differential strain to extract the monotonic decaying trend by using the time-domain and data-driven Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm. A linear regression scheme is then used to obtain the slope of the transformed non-linear trend, which carries information about the strain TC. Assessment of Vascular Permeability (VP), a transport parameter indicative of tumor growth, requires accurate strain TC estimations. Finite Element (FE), ultrasound simulations and in vivo experiments are used to investigate the performance of the proposed technique. Based on the simulation analysis, the average Percentage Relative Error (PRE) values of our method are 4.15% (for TC estimation) and 5.00% (for VP estimation) at 20 dB SNR level for different Percentage of Good Frames (PGF) (i.e., 20%, 50%, 75%, and 100%). These PRE values are substantially lower than those obtained using other conventional elastographic techniques. Our proposed method could become a new data-adaptive tool for analyzing the non-linear time-dependent response of complex tissues such as cancers.

Dynamic Particle Filter Framework for Robust Object Tracking

Most of siamese network and correlation filter (CF) based trackers usually employ the context regression scheme to achieve appealing performance in both accuracy and efficiency. However, they are prone to drifting in challenging situations exhibiting occlusion, out-of-view and large-scale variations due to the lack of failure correction ability in these regressors. Particle filter based trackers can help to recover from tracking failures since several particles of high confidence about the target can be remained for the probability estimation of next frames, but need the large numbers of particles for each frame. In this paper, we propose a generic dynamic particle filter framework, which can reasonably control the number of particles in different scenarios, to improve the robustness of siamese and CF trackers by jointing the target classifier to relieve drifting. Our fundamental insight is that general scenarios are processed efficiently by the context regressor with few particles, while special scenarios are handled effectively by the target classifier with many particles. We propose a novel measure to determine whether to adopt the regressor and few particles to estimate target states with high measure scores, or increase the number of particles to prevent drifting with the proposed multi-template matching strategy in the classifier. In extensive experiments on eight large-scale benchmarks including OTB, UAV, TC128, VOT2017, VOT2019, LaSOT, TrackingNet and Got-10k, the proposed framework enables many basic siamese and CF trackers to operate at least over 24 frames per second and achieve superior tracking performance than themself, as well as the comparable accuracy with the state-of-the-art trackers. Furthermore, our framework is flexible and still has great potential for improvement and generalization.

Pharmaceutical regulation and policies in Austria

Austria is a middle-sized, high-income country in Central Europe with universal health coverage and a highly fragmented health care system. Medicines for outpatient use that are reimbursed by the social health insurance are price-regulated, whereas medicines for inpatient use are neither subject to price regulation nor to health technology assessment (HTA). These medicines are procured by hospitals, hospital groups and provinces which are the main owner of public hospitals. The major pricing policy for new medicines in the outpatient sector is external price referencing. Austria refers to all other 26 European Union (EU) Member States; the determined benchmark price must not exceed the EU average price. External price referencing is based on list prices, but the statutory manufacturer discounts applicable in some reference countries are considered. Regular price reviews with subsequent price adjustments are in place. For generic and biosimilar medicines to be included in the outpatient positive list, a price link policy is applied with different reduction rates for generics and biosimilars. In the supply chain, prices of all medicines are regulated through regressive mark-up schemes for wholesale and community pharmacies. The inclusion in the outpatient positive list is based on an HTA process which comprises pharmacological, medical-therapeutic and economic evaluations, followed by price negotiations about the reimbursement price. For medicines with high financial burden for the public payers, the Austrian Social Insurance (for the outpatient sector) and procurers for hospitals tend to conclude managed-entry agreements with confidential discounts. For outpatient medicines, patients are charged a fixed prescription fee per prescribed item; no further co-payments apply for outpatient or inpatient medicines. Studies have shown that the current pricing and reimbursement policies in Austria have contributed to keep prices of outpatient medicines stable (high-cost medicines tend to range above EU average), whereas the unregulated prices of medicines used in hospitals are frequently the highest in European comparison. Pharmacy mark-ups are also high in comparison to other European countries.

An algorithmic framework for computational estimation of soil freezing characteristic curves

AbstractMany numerical models for simulating freezing and thawing phenomena of soil have been developed due to emerging geotechnical issues in cold regions. In particular, coupled thermo‐hydro‐mechanical (THM) analysis is used to evaluate complicated deformation, thermal, and moisture transport behavior of freezing–thawing soils. This study proposes a soil‐freezing characteristic curve (SFCC) that is robust and adaptive with various computational frameworks, including the THM approach. The proposed SFCC can also account for different soil types by incorporating the particle size distribution. Here an automatic regression scheme is adopted to update the SFCC associated with deformation and thermal changes. In addition, a smoothing algorithm is adopted to prevent a sharp change of the SFCC due to phase transition between the liquid water and crystal ice. Based on experimental works in the literature, the applicability of our model is demonstrated when the initial water contents and soil particle distribution differ. We further investigate the performance of the proposed SFCC as a constitutive model within a simplified THM framework. Our results show that the proposed model captures the desired behavior of different soil types in the freezing process, such as freezing temperature depreciation, the effect of compaction, and mechanical loading on unfrozen water content.

Verifiable Data Mining Against Malicious Adversaries in Industrial Internet of Things

With the large-scaled data generated from various interconnected machines and networks, Industrial Internet of Things (IIoT) provides unprecedented opportunities for facilitating data mining for industrial applications. The current IIoT architecture tends to adopt cloud computing for further timely mining IIoT data, however, the openness of security-critical IIoT becomes challenging in terms of unbearable privacy issues. Most existing privacy-preserving data mining (PPDM) techniques are designed to resist honest-but-curious adversaries (i.e., cloud servers and data users). Due to the complexity and openness in IIoT, PPDM is significantly difficult with the presence of malicious adversaries in IIoT who may incur incorrect learned models and inference results. To solve the aforementioned issues, we propose a framework to extend existing PPDM to guard linear regression against malicious behaviors (hereafter referred to as GuardLR). To prevent dishonest computations of cloud servers and inconsistent inputs of data users, we first design a privacy-preserving verifiable learning scheme for linear regression, which guarantees the correctness of learning. In this article, to avoid malicious clouds from returning incorrect inference results, we design a privacy-preserving prediction scheme with lightweight verification. Our formal security analysis shows that GuardLR achieves privacy, completeness, and soundness. Empirical experiments using real-world datasets also demonstrate that GuardLR has high computational efficiency and accuracy.

Super-resolution Image Reconstruction Based on Double Regression Network Model

Abstract By learning nonlinear mapping functions from low resolution (LR) images to high resolution (HR) images, deep neural networks show good performance in image super-resolution (SR). However, the existing SR approach has two potential limitations. First, learning the mapping function from LR to HR images is usually an ill-conditioned problem, since there exist an infinite number of HR images that can be down-sampled to the same LR image. Thus, the space of possible functions can be very large, making it difficult to find a good solution. Second, paired LR-HR data may not be available in real-world applications, and the underlying degradation method is often unknown. For this more general case, existing SR models tend to generate adaptive problems and produce poor performance. To solve the above problem, we propose a dual regression scheme that reduces the space of possible functions by introducing additional constraints on LR data.

A Novel Two-Mode Integral Approach for Thermal Error Modeling in CNC Milling-Turning Machining Center

Thermal errors have the largest contribution, as much as about 70%, to the machining inaccuracy of computer-numerical-controlled (CNC) machining centers. The error compensation method so far plays the most popular and effective way to minimize the thermal error. How to accurately and quickly build an applicable thermal error model (TEM) is the kernel work of thermal error compensation. On the basis of some comprehensive machine-learning schemes, past proposed TEMs had impressive merits for dealing with the thermal error modeling of single-function (milling or turning cutting) machine tools with only considering one set of thermal key points. These proposed modelling methodologies become worse when applied to CNC compound milling-turning machining centers in actual cutting applications. This paper proposes a two-mode integral TEM based on the Lasso and the random forest regression schemes to quickly and accurately predict the thermal deformations of such a machine. The first mode is the thermal error modeling for milling cutting conditions, and the second mode is that for turning cutting conditions. For data reduction, two different sets of temperature key points, one for milling and the other for turning, are obtained. Then, on the basis of the random forest regression scheme, we separately establish two TEMs but concurrently use them to predict the tool-center-point deformations of both milling as well as turning spindles. Further, we compare our proposed TEM with several frequently-used machine-learning-based TEMs and the results show that our proposed TEM are the best among all, no matter in the modelling experiment or in the test experiment. The proposed TEM has a maximum prediction error of 6.08 &#x03BC;<i>m</i> for milling cutting and that of 1.455&#x03BC;<i>m</i> for turning cutting in the modeling experiment. By our proposed two-mode integral TEM, the thermal error of a multi-function milling-turning machining center can be accurately predicted and quickly compensated.

Sparsest Univariate Learning Models Under Lipschitz Constraint

Beside the minimizationof the prediction error, two of the most desirable properties of a regression scheme are <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">stability</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">interpretability</i> . Driven by these principles, we propose continuous-domain formulations for one-dimensional regression problems. In our first approach, we use the Lipschitz constant as a regularizer, which results in an implicit tuning of the overall robustness of the learned mapping. In our second approach, we control the Lipschitz constant explicitly using a user-defined upper-bound and make use of a sparsity-promoting regularizer to favor simpler (and, hence, more interpretable) solutions. The theoretical study of the latter formulation is motivated in part by its equivalence, which we prove, with the training of a Lipschitz-constrained two-layer univariate neural network with rectified linear unit (ReLU) activations and weight decay. By proving representer theorems, we show that both problems admit global minimizers that are continuous and piecewise-linear (CPWL) functions. Moreover, we propose efficient algorithms that find the sparsest solution of each problem: the CPWL mapping with the least number of linear regions. Finally, we illustrate numerically the outcome of our formulations.

Multi-fidelity regression using artificial neural networks: Efficient approximation of parameter-dependent output quantities

Highly accurate numerical or physical experiments are often very time-consuming or expensive to obtain. When time or budget restrictions prohibit the generation of additional data, the amount of available samples may be too limited to provide satisfactory model results. Multi-fidelity methods deal with such problems by incorporating information from other sources, which are ideally well-correlated with the high-fidelity data, but can be obtained at a lower cost. By leveraging correlations between different data sets, multi-fidelity methods often yield superior generalization when compared to models based solely on a small amount of high-fidelity data. In the current work, we present the use of artificial neural networks applied to multi-fidelity regression problems. By elaborating a few existing approaches, we propose new neural network architectures for multi-fidelity regression. The introduced models are compared against a traditional multi-fidelity regression scheme — co-kriging. A collection of artificial benchmarks are presented to measure the performance of the analyzed models. The results show that cross-validation in combination with Bayesian optimization leads to neural network models that outperform the co-kriging scheme. Additionally, we show an application of multi-fidelity regression to an engineering problem. The propagation of a pressure wave into an acoustic horn with parametrized shape and frequency is considered, and the index of reflection intensity is approximated using the proposed multi-fidelity models. A finite element, full-order model and a reduced-order model built through the reduced basis method are adopted as the high- and low-fidelity, respectively. It is shown that the multi-fidelity neural networks return outputs that achieve a comparable accuracy to those from the expensive, full-order model, using only very few full-order evaluations combined with a larger amount of inaccurate but cheap evaluations of the reduced order model.

Distributed sparsity-based non-linear regression with multiple kernels in wireless sensor networks

In this paper, we investigate the problem of non-linear time-varying regression in a wireless sensor network system. A field over an area is estimated using a number of fixed sensors distributed randomly in the region. Two regression schemes have been suggested: one for centralized regression and the other for distributed regression. The proposed method exploits the reweighted multiple kernel least squares as the data fidelity term. Moreover, using the idea that not all of the sensors affect the field value in a specific region point, we add the sparsity constraint on the weighting vector to obtain a better estimation of the field. Another TV-norm term is added to the total cost function to ensure the smoothness of the estimated field. The consensus-based term is adopted to develop a distributed estimation technique. In addition, the bandwidth of the kernels are defined to change adaptively over time using a least mean square (LMS) technique which enables the algorithm to track the time-variations of the under-lying field. The simulation results confirm the superiority of the proposed method over its counterparts in both synthetic and real data.

Through the eyes of a descriptor: Constructing complete, invertible descriptions of atomic environments

In this work we apply methods for describing three-dimensional images to the problem of encoding atomic environments in a way that is invariant to rotations, translations, and permutations of the atoms and, crucially, can be decoded back into the original environment modulo global orientation without the need for training a model. From the point of view of decoding, the descriptor is optimally complete and can be extended to arbitrary order, allowing for a systematic convergence of the fidelity of the description. In experiments on molecules ranging from 3 to 29 atoms in size, we demonstrate that positions can be decoded with a 97% success rate and positions plus species with a 70% rate of success, rising to 95% if a second fingerprint is used. In all cases, consistent recovery is observed for molecules with 17 or fewer atoms. Additionally, we evaluate the descriptor's performance in predicting the energies and forces of bulk Ni, Cu, Li, Mo, Si, and Ge by means of a neural network model trained on DFT data. When comparing to six machine learning interaction potential methods that use various descriptors and regression schemes, our descriptor is found to be competitive, in several cases outperforming well established methods. The combined ability to both decode and make property predictions from a representation that does not need to be learned lays the foundations for a novel way of building generative models that are tasked with solving the inverse problem of predicting atomic arrangements that are statistically likely to have certain desired properties.

Cellular Traffic Prediction via a Deep Multi-Reservoir Regression Learning Network for Multi-Access Edge Computing

Cellular traffic prediction at mobile edges is extremely valuable to ultra high-reliability low-latency (URLLC) communication of 5G. Many network actions depend on this prediction technology, ranging from radio resource scheduling, edge node sharing, and traffic control, to network slicing and dynamic network function virtualization. However, accurate prediction for cellular traffic flow is a tough challenge, since irregular and dramatic fluctuations of network traffic, caused by continuous topology update and multifarious service requests, occur frequently. Motivated by these investigations, we propose a deep multi-reservoir regression learning network for cellular traffic prediction, called mRDLN, supporting multi-access edge computing. This architecture is a uniform and consistent system with functional modules of smoothing, feature extraction, and multi-reservoir regression. This is the first attempt to enhance deep neural computing considering a combined regression scheme in the framework of a deep belief network, where multiple echo state networks are integrated to perform local supervised learning instead of the original single approximator. On highly bursty cellular traffic traces, experimental simulations show that our mRDLN consistently outperforms the state-of-the-art models, and the superior prediction is further verified by means of statistical analysis.