Reduction In Computational Complexity Research Articles

An innovative model for electrical load forecasting: A case study in Australia

Short-term load prediction has always played an increasingly important part in power system administration, load dispatch, and energy transfer scheduling. However, how to build a novel model to improve the accuracy of load forecasts is not only an extremely challenging problem but also a concerning problem for the power market. Specifically, the individual model pays no attention to the significance of data selection, data preprocessing, and model optimization. So these models cannot always satisfy the time series forecasting’s requirements. With these above-mentioned ignored factors considered, to enhance prediction accuracy and reduce computation complexity, in this study, a novel and robust method were proposed for multi-step forecasting, which combines the power of data selection, data preprocessing, artificial neural network, rolling mechanism, and artificial intelligence optimization algorithm. Case studies of electricity power data from New South Wales, Australia, are regarded as exemplifications to estimate the performance of the developed novel model. The experimental results demonstrate that the proposed model has significantly increased the accuracy of load prediction in all quarters. As a result, the proposed method not only is simple, but also capable of achieving significant improvement as compared with the other forecasting models, and can be an effective tool for power load forecasting.

Single-Level Differentiable Contact Simulation

We present a differentiable formulation of rigid-body contact dynamics for objects and robots represented as compositions of convex primitives. Classical physics engines rely on non-differentiable collision detection modules. More recent optimization-based approaches simulating contact between convex primitives rely on a bilevel formulation that separates collision detection and contact simulation. These latest approaches are unreliable in realistic contact simulation scenarios because isolating the collision detection problem introduces contact location non-uniqueness. Our approach combines contact simulation and collision detection into a unified single-level optimization problem. This disambiguates the collision detection problem in a physics-informed manner. Our formulation features improved simulation robustness and a reduction in computational complexity when compared to a similar <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">differentiable</i> simulation baseline. We illustrate the contact and collision differentiability on a robotic manipulation task requiring optimization-through-contact. We provide a numerically efficient implementation of our formulation called <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/simon-lc/Silico.jl</uri> Silico.jl.

Support Vector Regression for Bluetooth Ranging in Multipath Environments

Ranging solutions for Internet of Things (IoT) localization applications seek to provide high accuracy with low cost of implementation. Among candidate IoT technologies that may fit this criterion, Bluetooth is a desirable choice as Bluetooth Low Energy (BLE) support is ubiquitous in modern smartphones, providing low implementation costs and low power consumption. Recent advancements in BLE ranging technology employ the Multi-Carrier Phase Difference technique which takes two-way channel frequency response (CFR) measurements. However, accurate ranging with these measurements is challenging due to many closely spaced multipath components from squaring the one-way CFR, a single or low number of snapshots, and model imperfections that arise in practical scenarios. To overcome these challenges, we propose a data-driven support vector regression (SVR) approach. Using real-world BLE measurements, our proposed SVR method demonstrates decimeter-level accuracy with single antenna devices, whereas Multiple Signal Classification (MUSIC), a popular model-based method, requires multiple antennas to obtain comparable performance. Moreover, we show robustness in different multipath environments including indoor, outdoor, and non-line-of-sight conditions, we determine generalization capabilities with training size, and we analytically establish the reduction in computational complexity compared to MUSIC.

BATFormer: Towards Boundary-Aware Lightweight Transformer for Efficient Medical Image Segmentation.

Transformers, born to remedy the inadequate receptive fields of CNNs, have drawn explosive attention recently. However, the daunting computational complexity of global representation learning, together with rigid window partitioning, hinders their deployment in medical image segmentation. This work aims to address the above two issues in transformers for better medical image segmentation. We propose a boundary-aware lightweight transformer (BATFormer) that can build cross-scale global interaction with lower computational complexity and generate windows flexibly under the guidance of entropy. Specifically, to fully explore the benefits of transformers in long-range dependency establishment, a cross-scale global transformer (CGT) module is introduced to jointly utilize multiple small-scale feature maps for richer global features with lower computational complexity. Given the importance of shape modeling in medical image segmentation, a boundary-aware local transformer (BLT) module is constructed. Different from rigid window partitioning in vanilla transformers which would produce boundary distortion, BLT adopts an adaptive window partitioning scheme under the guidance of entropy for both computational complexity reduction and shape preservation. BATFormer achieves the best performance in Dice of 92.84 %, 91.97 %, 90.26 %, and 96.30 % for the average, right ventricle, myocardium, and left ventricle respectively on the ACDC dataset and the best performance in Dice, IoU, and ACC of 90.76 %, 84.64 %, and 96.76 % respectively on the ISIC 2018 dataset. More importantly, BATFormer requires the least amount of model parameters and the lowest computational complexity compared to the state-of-the-art approaches. Our results demonstrate the necessity of developing customized transformers for efficient and better medical image segmentation. We believe the design of BATFormer is inspiring and extendable to other applications/frameworks.

A Digital Watermark Method for In-Vehicle Network Security Enhancement

In recent years, automobiles have become increasingly networked with the development of information and communication technology (ICT). With the enhancement of the connectivity between the electronic control units of the vehicle and the external network, cybersecurity and driving safety issues have also arisen. However, as a traditional in-vehicle network standard, the controller area network (CAN) protocol lacks a network security mechanism design, making it vulnerable to hacker attacks. The current research on the cybersecurity enhancement method for in-vehicle CAN is a multi-objective optimization problem with response time, bandwidth consumption and computational complexity reduction. This study presents an in-vehicle message authentication method based on digital watermark and Huffman coding that are compatible with current CAN bus protocols. Our method not only minimize the delay of message authentication, but also save the storage resources to the greatest extent. Moreover, to tackle the issue of low computing resources, multiple messages can be authenticated at one time. In order to evaluate the performance, we implement our method on CAN communication platform and validate its high accuracy, low latency, and compatibility.

Lightweight Machine Learning for Digital Cross-Link Interference Cancellation With RF Chain Characteristics in Flexible Duplex MIMO Systems

The flexible duplex (FD) technique, including dynamic time-division duplex (D-TDD) and dynamic frequency-division duplex (D-FDD), is regarded as a promising solution to achieving a more flexible uplink/downlink transmission in 5G-Advanced or 6G mobile communication systems. However, it may introduce serious cross-link interference (CLI). For better mitigating the impact of CLI, we first present a more realistic base station (BS)-to-BS channel model incorporating the radio frequency (RF) chain characteristics, which exhibit a hardware-dependent nonlinear property, and hence the accuracy of conventional channel modelling is inadequate for CLI cancellation. Then, we propose a channel parameter estimation based polynomial CLI canceller and two machine learning (ML) based CLI cancellers that use the lightweight feedforward neural network (FNN). Our simulation results and analysis show that the ML based CLI cancellers achieve notable performance improvement and dramatic reduction of computational complexity, in comparison with the polynomial CLI canceller.

IDDANet: An Input-Driven Dynamic Adaptive Network ensemble method for edge intelligence

Dynamic pruning and conditional convolution are crucial for reducing the computational complexity of convolutional neural networks, enabling their deployment on resource-limited devices. However, existing methods only focus on one aspect in isolation, ignoring the collaborative potential of devices and the similarity between input samples in edge cluster scenarios, making it difficult to eliminate redundancy in a given network structure. In order to reduce computation costs while enhancing model representation, we propose an input-driven dynamic adaptive network ensemble method, namely IDDANet. Our method is a two-stage model synthesis strategy that first dynamically prunes the network by using sample complexity and similarity to obtain a lightweight model and a set of input-dependent combination coefficients. We then combine the convolution parameters of multiple models to synthesize a more accurate ensemble model. As demonstrated through extensive experiments on different datasets, our method can significantly reduce computation complexity while improving model representation ability compared to state-of-the-art methods. For example, our method can reduce 58.5% of the FLOPs of ResNet-18, while only having a 0.1% Top-1 accuracy drop on CIFAR-10.

Attention guided domain alignment for conditional face image generation

Despite the great success of Generative Adversarial Network (GAN) in face image generation, it is still a challenge to generate a faithful yet high-fidelity face image given an exemplar image and a conditional input from a distinct domain (e.g., a semantic segmentation mask or an edge map). Existing methods learn image-level features to align distinct domains in an intermediate domain, which ignore the spatial relationship of the facial semantic parts and therefore lead to semantic mismatching problems. In addition, it is computationally expensive to establish dense correspondences in the intermediate domain, especially for high-resolution face images. To address these problems, we propose a novel attention guided domain alignment method for conditional face image generation, which aligns two domains directly under the guidance of the local attention learned from semantically similar face parts. In particular, we assign a dedicated index for each feature block and adopt a top-k ranking operation to correspond block-wise features in two distinct domains, which exploits the spatial relationship of facial parts and preserves the texture structure during alignment. The local attention is then learned from the retrieved blocks, which reduces computation complexity substantially and therefore enables to build alignment in a high resolution. The aligned features are finally fused with adaptive weights learned from their long-range correlation coefficients, which capture the semantic coherence of the style features between the two domains. Extensive experimental results on the CelebAMask-HQ dataset demonstrate that the proposed method is superior to the state-of-the-art methods.

Adaptive Fuzzy Logic Deep-Learning Equalizer for Mitigating Linear and Nonlinear Distortions in Underwater Visible Light Communication Systems

Underwater visible light communication (UVLC) has recently come to light as a viable wireless carrier for signal transmission in risky, uncharted, and delicate aquatic environments like seas. Despite the potential of UVLC as a green, clean, and safe alternative to conventional communication methods, it is challenged by significant signal attenuation and turbulent channel conditions compared to long-distance terrestrial communication. To address linear and nonlinear impairments in UVLC systems, this paper presents an adaptive fuzzy logic deep-learning equalizer (AFL-DLE) for 64 Quadrature Amplitude Modulation-Component minimal Amplitude Phase shift (QAM-CAP)-modulated UVLC systems. The proposed AFL-DLE is dependent on complex-valued neural networks and constellation partitioning schemes and utilizes the Enhanced Chaotic Sparrow Search Optimization Algorithm (ECSSOA) to improve overall system performance. Experimental outcomes demonstrate that the suggested equalizer achieves significant reductions in bit error rate (55%), distortion rate (45%), computational complexity (48%), and computation cost (75%) while maintaining a high transmission rate (99%). This approach enables the development of high-speed UVLC systems capable of processing data online, thereby advancing state-of-the-art underwater communication.

Improving the accuracy and speed of fast template‐matching algorithms by neural architecture search

AbstractNeural architecture search can be used to find convolutional neural architectures that are precise and robust while enjoying enough speed for industrial image processing applications. In this paper, our goal is to achieve optimal convolutional neural networks (CNNs) for multiple‐templates matching for applications such as licence plates detection (LPD). We perform an iterative local neural architecture search for the models with minimum validation error as well as low computational cost from our search space of about 32 billion models. We describe the findings of the experience and discuss the specifications of the final optimal architectures. About 20‐times error reduction and 6‐times computational complexity reduction is achieved over our engineered neural architecture after about 500 neural architecture evaluation (in about 10 h). The typical speed of our final model is comparable to classic template matching algorithms while performing more robust and multiple‐template matching with different scales.

Multi-band chaotic non-orthogonal matrix-based encryption for physical-layer security enhancement in OFDM-PONs

In this paper, we propose and experimentally demonstrate a novel multi-band chaotic non-orthogonal matrix (CNOM)-based encryption scheme for secure orthogonal frequency division multiplexing (OFDM) passive optical networks. The dimension of non-orthogonality is exploited in the encryption with the CNOM, where both faster-than-Nyquist signaling and redundant precoding were employed to dynamically scramble the original number of subcarriers of each sub-band and fix the overall data rate. Both simulation and experimental results of a 10.6 Gb/s 4-QAM transmission over a 20-km standard single-mode fiber showed that the total key space significantly increases by 1.4×10482 and 1.72×10653 times, as the number of sub-bands increases from 1 to 10 and 15, with a considerable reduction in computational complexity of 90% and 93.33% in complex-valued multiplication, and 90.76% and 94.12% in complex-valued addition, when encrypting one OFDM symbol, respectively. Moreover, the improved resilience to the frequency roll-off has also been verified by using additional permutation matrices in the proposed encryption algorithm.

Ensemble Learning Aided QPSO-Based Framework for Secrecy Energy Efficiency in FD CR-NOMA Systems

Cognitive radio (CR), non-orthogonal multiple access (NOMA), and full-duplex (FD) communications have been considered key technologies for providing spectrum utilization improvement and higher energy efficiency on the internet of things (IoT) networks and next-generation communication. However, security concerns are still an issue to be addressed because confidential information is exposed in wireless systems. To solve this problem, we design a novel artificial intelligence (AI)-based framework for maximizing the secrecy energy efficiency (SEE) in FD cooperative relay underlay CR-NOMA systems that are exposed to multiple eavesdroppers. First, we formulate the non-convex SEE optimization problem as bi-level optimization, subject to constraints that satisfy the quality-of-service requirements of secondary users. In particular, the outer problem is solved with ensemble learning (EL) to select the optimal relay. Regarding the inner problem, we propose a quantum particle swarm optimization (QPSO)-based technique to optimize power allocation. In addition, for comparison purposes, we describe a cooperative relay CR network with orthogonal multiple access (OMA), rate-splitting multiple access (RSMA), and half-duplex technologies. Moreover, we evaluate comparative schemes based on machine learning algorithms and swarm intelligence baseline schemes. Furthermore, the proposed EL-aided QPSO-based framework achieves performance close to the optimal solutions, with a meaningful reduction in computation complexity.

Adaptive Channel Assignment for Maneuvering Target Tracking in Multistatic Passive Radar

In this paper, two adaptive channel assignment (CA) schemes are proposed for maneuvering target tracking (MTT) in multistatic passive radar. For a predetermined total number of channels w.r.t. each receiver, (a) the first scheme selects <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$Q$</tex-math></inline-formula> channels to achieve the highest MTT accuracy; (b) the second scheme minimizes the number of channels while guaranteeing a required MTT performance. By utilizing the best-fitting Gaussian approximation, we derive the predicted conditional Cramér-Rao lower bound to evaluate the impact of CA on MTT performance, and formulate CA schemes as convex integer program problems, since channel variables are in binary form. The objective in the first CA problem is smooth, so we apply the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$l_{2}$</tex-math></inline-formula> -box alternating direction method of multipliers to divide it into several simple subproblems and alternatively solve them in continuous domains without the introduction of relaxation error. The second CA problem may be infeasible when the required accuracy cannot be met with available channels. To tackle it, we introduce a linear rectification function w.r.t the performance constraint, and equivalently reformulate it as a feasible one. Then, a fast multi-start add-one-channel policy is designed to minimize the gap between the actual accuracy and the accuracy requirement. These algorithms offer considerable reductions in computational complexity compared with the branch and bound method and may achieve near-optimal performance. Simulation results demonstrate that these schemes can either achieve the highest MTT accuracy with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$Q$</tex-math></inline-formula> channels or reduce the number of channels while guaranteeing MTT performance, compared with the traditional random channel allocation scheme.

Antenna Beampattern Matched Optimum Coherent Detection in High-Resolution Mechanically Scanning Maritime Surveillance Radars

Mechanically scanning radars are still widely used in ocean guard and surveillance areas nowadays, owing to simple structure, light weight and low cost. The antenna rotation causes amplitude modulation of target returns along pulses by the antenna azimuth beampattern. The traditional coherent detection that ignores this modulation suffers from some performance loss. In this paper, an antenna beampattern matched optimum coherent detector and a dual-threshold detection scheme are proposed for high-resolution mechanically scanning maritime surveillance radars. Firstly, it is shown that the modulation of target returns is determined by the antenna azimuth beampattern, scan rate, and pulse repetition interval (PRI) and the antenna scanning also affect the temporal correlation of sea clutter. Secondly, the antenna beampattern matched optimum coherent detector is given in compound-Gaussian sea clutter with inverse Gamma distributed texture and the pulse scanning mode is used to deal with unknown target azimuth angle in the matched filter. Thirdly, to mitigate the high computational cost of the pulse scanning mode, a dual-threshold scheme is given, which searches all pulse-range cells by short-time integration and further confirms suspected cells whose test statistics are between the two lower and higher thresholds in the search by long-time integration. Moreover, two fast recursive clutter covariance matrix inversion algorithms along pulses and range cells are presented, which bring a great reduction in computational complexity. Finally, real measured radar data is used to verify the proposed detector and scheme and the experimental results show that it behaves better in detection of small targets than the traditional detectors.

Effective implementation and improvement of fast labeled multi-Bernoulli filter

Effective implementation of the fast labeled multi-Bernoulli (FLMB) filter is addressed for target tracking with interval measurements. Firstly, a sequential Monte Carlo (SMC) implementation of the FLMB filter, SMC-FLMB filter, is derived based on generalized likelihood function weighting. Then, a box particle (BP) implementation of the FLMB filter, BP-FLMB filter, is developed, with a computational complexity reduction of the SMC-FLMB filter. Finally, an improved version of the BP-FLMB filter, improved BP-FLMB (IBP-FLMB) filter, is proposed, improving its estimation accuracy and real-time performance under the conditions of low detection probability and high clutter. Simulation results show that the BP-FLMB filter has a great improvement of the real-time performance than the SMC-FLMB filter, with similar tracking performance. Compared with the BP-FLMB filter, the IBP-FLMB filter has better estimation performance and real-time performance under the conditions of low detection probability and high clutter.

A Novel Low-Complexity Method for Near-Field Sources Based on an S-IMISC Array Model

Array optimization has recently received significant attention owing to its several advantages, such as larger array aperture and greater degrees of freedom (DOFs). However, current works focus on far-field sources, while array optimization for near-field sources has not been adequately investigated. Therefore, this work develops a new symmetry sparse array model for near-field sources based on the improved maximum inter-element spacing constraint (IMISC). The proposed symmetry IMISC (S-IMISC) array model has all the advantages of traditional sparse array models. Compared with traditional sparse array models, the S-IMISC array model affords more uniform DOFs and is less affected by mutual coupling. Additionally, in order to improve the real-time performance of near-field sources localization, the characteristic equation-based method (CEM) is used to obtain the azimuth information of near-field sources which can avoid eigenvalue decomposition (EVD), and a spectrum peak search and compression scheme is used to obtain the distance information by searching the partial area instead of the whole Fresnel area, thereby significantly reducing computation complexity. Extensive simulations verify the advantages of the proposed algorithm and the S-IMISC array model.

Deep Learning-Based Symptomizing Cyber Threats Using Adaptive 5G Shared Slice Security Approaches

In fifth Generation (5G) networks, protection from internal attacks, external breaches, violation of confidentiality, and misuse of network vulnerabilities is a challenging task. Various approaches, especially deep-learning (DL) prototypes, have been adopted in order to counter such challenges. For 5G network defense, DL module are recommended here in order to symptomize suspicious NetFlow data. This module behaves as a virtual network function (VNF) and is placed along a 5G network. The DL module as a cyber threat-symptomizing (CTS) unit acts as a virtual security scanner along the 5G network data analytic function (NWDAF) to monitor the network data. When the data were found to be suspicious, causing network bottlenecks and let-downs of end-user services, they were labeled as “Anomalous”. For the best proactive and adaptive cyber defense system (PACDS), a logically organized modular approach has been followed to design the DL security module. In the application context, improvements have been made to input features dimension and computational complexity reduction with better response times and accuracy in outlier detection. Moreover, key performance indicators (KPIs) have been proposed for security module placement to secure interslice and intraslice communication channels from any internal or external attacks, also suggesting an adaptive defense mechanism and indicating its placement on a 5G network. Among the chosen DL models, the CNN model behaves as a stable model during behavior analysis in the results. The model classifies botnet-labeled data with 99.74% accuracy and higher precision.

Deep convolutional architecture-based hybrid learning for sleep arousal events detection through single-lead EEG signals.

Detecting arousal events during sleep is a challenging, time-consuming, and costly process that requires neurology knowledge. Even though similar automated systems detect sleep stages exclusively, early detection of sleep events can assist in identifying neuropathology progression. An efficient hybrid deep learning method to identify and evaluate arousal events is presented in this paper using only single-lead electroencephalography (EEG) signals for the first time. Using the proposed architecture, which incorporates Inception-ResNet-v2 learning transfer models and optimized support vector machine (SVM) with the radial basis function (RBF) kernel, it is possible to classify with a minimum error level of less than 8%. In addition to maintaining accuracy, the Inception module and ResNet have led to significant reductions in computational complexity for the detection of arousal events in EEG signals. Moreover, in order to improve the classification performance of the SVM, the grey wolf algorithm (GWO) has optimized its kernel parameters. This method has been validated using pre-processed samples from the 2018 Challenge Physiobank sleep dataset. In addition to reducing computational complexity, the results of this method show that different parts of feature extraction and classification are effective at identifying sleep disorders. The proposed model detects sleep arousal events with an average accuracy of 93.82%. With the lead present in the identification, the method becomes less aggressive in recording people's EEG signals. According to this study, the suggested strategy is effective in detecting arousals in sleep disorder clinical trials and may be used in sleep disorder detection clinics.

Efficient segment-update block LMS-Newton algorithm for active control of road noise

For active control of road noise inside a vehicle, many reference signals are required in adaptive control systems, resulting in slow convergence speeds and heavy computational loads. In this paper, a means of reducing the computational complexity of adaptive control algorithms is investigated based on the fast-converging least-mean-squares Newton (LMS-Newton) algorithm. By analyzing the autocorrelation matrix of reference signals filtered by secondary paths, it is found that the dimensions of the autocorrelation matrix of filtered reference signals do not necessarily have to equal the dimensions of the adaptive control filters, and instead the dimensions can be determined from the correlation between reference signals filtered by secondary paths. Then a computationally efficient segment-update block algorithm is proposed that splits adaptive control filters into several segments and the coefficients of each segment are updated independently. Theoretical analysis shows that the proposed segment-update block algorithm has similar convergence speed and noise-reduction performance as the traditional LMS-Newton algorithm when a segment of appropriate length is chosen, but the computational complexity is reduced dramatically because the inverse autocorrelation matrix is replaced by a smaller matrix. The performance of the proposed algorithm is verified through simulation experiments with different recorded road noises and secondary paths, and the results are compared with those of some typical algorithms. The simulation results showed that the proposed algorithm achieved a similar noise-reduction performance with an approximately 8.9-times reduction of computational complexity compared with the traditional LMS-Newton algorithm.

Self-Paced Dual-Axis Attention Fusion Network for Retinal Vessel Segmentation

The segmentation of retinal vessels plays an essential role in the early recognition of ophthalmic diseases in clinics. Increasingly, approaches based on deep learning have been pushing vessel segmentation performance, yet it is still a challenging problem due to the complex structure of retinal vessels and the lack of precisely labeled samples. In this paper, we propose a self-paced dual-axis attention fusion network (SPDAA-Net). Firstly, a self-paced learning mechanism using a query-by-committee algorithm is designed to guide the model to learn from easy to hard, which makes model training more intelligent. Secondly, during fusing of multi-scale features, a dual-axis attention mechanism composed of height and width attention is developed to perceive the object, which brings in long-range dependencies while reducing computation complexity. Furthermore, CutMix data augmentation is applied to increase the generalization of the model, enhance the recognition ability of global and local features, and ultimately boost accuracy. We implement comprehensive experiments validating that our SPDAA-Net obtains remarkable performance on both the public DRIVE and CHASE-DB1 datasets.