Cognitive Computation and Systems最新文献

Fixed control parameters and various optimisation objectives significantly limit the robot control performance. To address such issues, a parameter adaptive controller based on deep reinforcement learning is introduced firstly to adjust control parameters according to the real-time system state. Further, multiple evaluation mechanisms are constructed to take account of optimisation objectives so that the controller can adapt to different control performance indexes by different evaluation mechanisms. Finally, the target pedestrian tracking control with mobile robots is selected as the validation case study, and the Proportional-Derivative Controller is chosen as the foundation controller. Several simulation and experimental examples are designed, and the results demonstrate that the proposed method shows satisfactory performance while taking account of multiple optimisation objectives.

An investigation on the effect of mental activity in quality perception is presented using simultaneous measurement of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS), in a subject-independent approach. Building a subject-independent model is a harder problem due to noise and high EEG variability between individuals, correlated components analysis (CorrCA) have been proposed to extract significant correlated components for a single subject that experiences multiple identical trials; this is done by identifying spatio-temporal patterns of activity that are well preserved across trials. The aim is to build a model based on neurophysiological data to assess text-to-speech quality. In order to build a subject independent model, we extended the use of CorrCA such that it can be applied to the subject independent model. The authors used two preprocessing steps, namely the subject dependent and the stimulus dependent preprocessing. The second preprocessing used the denoising source separation (DSS) to remove noise/artefact that are subject specific. The discrete convolution is used for data fusion and the support vector machine for regression. With the proposed model, the fusion of EEG and fNIRS performs better than single modality. Using our defined regression accuracy metrics, the authors obtained accuracy of 81.346% for overall impression, 83.28% for valence and 89.714% for arousal. The model compete the baseline that is subject dependent.

Magnetic Resonance Imaging (MRI) based Autism Spectrum Disorder (ASD) detection approaches face various challenges due to variations in brain connectivity patterns, limited sample sizes, and heterogeneity of available data. These challenges make it hard to find consistent imaging markers. To address these issues, researchers have focused on advanced analysis methods, such as multi-modal imaging techniques and graph-based approaches to gain a comprehensive understanding of ASD neurobiology. However, existing graph-based approaches for ASD detection have primarily focused on pairwise similarities between individuals, neglecting individual characteristics and features. A novel framework to detect ASD using a Multi-Scale Enhanced Graph Convolutional Network (MSE-GCN). The framework combines the functional connectivity of resting-state functional MRI (rs-fMRI) with non-imaging phenotype data from Autism Brain Imaging Data Exchange-I (ABIDE-I). The framework uses MSE-GCN to represent individuals as node in a population graph. Each node corresponds to an individual and connects to feature vectors from imaging data. Edge weights between nodes are assigned to integrate phenotypic information. Then, the multiple parallel GCN layers are designed using random walk embedding. The output of these GCN layers is then combined in the fully connected layer to detect ASD effectively. The performance of the framework is evaluated using the ABIDE-I dataset. In addition, Recursive Feature Elimination and Multilayer Perceptron are utilised for feature selection. The outcome of this approach shows more than 10% advancement in accuracy, achieving an accuracy of 83% by incorporating phenotypic data in conjunction with MRI data within a GCN.

The pace of technological advancement is accelerating, and one of the latest developments is the emergence of Augmented Reality (AR) and Mixed Reality (MR) glasses as an extension of smartphones. The key to success lies in innovative research and technology that can reach a wide audience. To ensure a positive user experience, AR/MR glasses must offer interfaces that are easy to use, memorable, and leave a lasting impression. While Nielsen's heuristics are widely accepted as the standard for usability, it is clear that non-traditional applications require a rethinking of these heuristics to best suit their unique needs. A fresh usability heuristic for augmented and MR applications is designed by combining and modifying the existing models, such as Nielsen's 10 heuristics, Technology Acceptance Model, and Software Usability Measurement Inventory. The resulting framework incorporates 21 main heuristics and 60 sub heuristics. The 21 main heuristics are further grouped into the Norman's cognitive theory model based on the three levels of processing. The industry experts evaluated and validated the usability framework and established a higher level of effectiveness in identifying more usability problems compared with Nielsen's heuristics.

The ability to effectively classify human emotion states is critically important for human-computer or human-robot interactions. However, emotion classification with physiological signals is still a challenging problem due to the diversity of emotion expression and the characteristic differences in different modal signals. A novel learning-based network architecture is presented that can exploit four-modal physiological signals, electrocardiogram, electrodermal activity, electromyography, and blood volume pulse, and make a classification of emotion states. It features two kinds of attention modules, feature-level, and semantic-level, which drive the network to focus on the information-rich features by mimicking the human attention mechanism. The feature-level attention module encodes the rich information of each physiological signal. While the semantic-level attention module captures the semantic dependencies among modals. The performance of the designed network is evaluated with the open-source Wearable Stress and Affect Detection dataset. The developed emotion classification system achieves an accuracy of 83.88%. Results demonstrated that the proposed network could effectively process four-modal physiological signals and achieve high accuracy of emotion classification.

The utilisation of P2P tech in streaming media live broadcasting algorithm can effectively link the nodes in the endpoint and accelerate the establishment of efficient transmission channels between nodes, so as to ensure the data feasibility of video live broadcasting of terminal nodes. The simulation system of immersive virtual reality tech provides intelligent interactive experience and design framework. Its integration with P2P network streaming media live broadcast can bring users a stronger sense of experience, interaction and substitution, and provide high-quality streaming media live broadcast service. On account of this, the authors first analyse the principle and utilisation of peer-to-peer network tech, then study the streaming media live broadcast algorithm considering peer-to-peer network and the design of P2P-based streaming media live broadcast algorithm, and finally give the immersive virtual reality simulation of P2P-based streaming media live broadcast scheduling algorithm.

Accurate magnetic resonance imaging (MRI) segmentation plays a critical role in the diagnosis and treatment of demyelinating diseases. But the existing automatic segmentation methods are not suitable for the segmentation of demyelinating lesions with small lesion size, highly diffuse edges and complex boundary shapes. An improved model is proposed for demyelinating diseases MRI segmentation based on the U-shaped structure convolution neural networks (UNet). A context information weighting fusion (CIWF) module and a modified channel attention (MCA) module are developed and embedded in UNet to address the small lesion region and diffuse edge issues. The CIWF module can dynamically screen and fuse shallow and deep features at different stages, making the model pay more attention to small lesions. The MCA module enables the model to learn diverse features by adding weights to the channel, which helps in diffuse edge segmentation. Comparisons with many existing methods on real-world demyelinating disease MRI segmentation dataset show that our method achieve the highest Dice metric.

The artificial intelligence (AI) within the last decade has experienced a rapid development and has attained power to simulate human-thinking in various situations. When the deep neural networks (DNNs) are trained with huge dataset and high computational resources it can bring out great outcomes. But the learning process of DNN is very much complicated and time-consuming. In various circumstances, where there is a data-scarcity, the algorithms are not capable of learning tasks at a faster rate and perform nearer to that of human intelligence. With advancements in deep meta-learning in several research studies, this problem has been dealt. Meta-learning has outspread range of applications where the meta-data (data about data) of the either tasks, data or the models which were previously trained can be employed to optimise the learning. So in order to get an insight of all existing meta-learning approaches for DNN model optimisation, the authors performed survey introducing different meta-learning techniques and also the current optimisation-based approaches, their merits and open challenges. In this research, the Reptile meta-learning algorithm was chosen for the experiment. As Reptile uses first-order derivatives during optimisation process, hence making it feasible to solve optimisation problems. The authors achieved a 5% increase in accuracy with the proposed version of Reptile meta-learning algorithm.

Although Dynamic Movement Primitives (DMP) is an effective tool for robotic arm trajectory generalisation, the application of DMP in the 3C (Computer, Communication, Consumer Electronics) industry still faces challenges such as low precision and high-time consumption. To address this problem, we propose a novel Cauchy DMP framework. The main improvements and advantages of Cauchy DMP, compared to the original DMP, are (1) since the Cauchy distribution has a simpler model and wider shape, using the Cauchy distribution instead of the Gaussian distribution in the original DMP reduces the complexity of the algorithm and saves time. (2) Singular Value Decomposition (SVD) can effectively model the error. To reduce the interference of the rounding and human error on the trajectory, SVD can be used to obtain the weight of each basis function. The proposed Cauchy DMP framework combines the above two points and is validated on a real UR5 robotic arm. The results show that Cauchy DMP retains the learnability of the original DMP and has the advantages of short time consumption and low error rate.

In this study, to address the issues of sluggish convergence and poor learning efficiency at the initial stages of training, the authors improve and optimise the Deep Deterministic Policy Gradient (DDPG) algorithm. First, inspired by the Artificial Potential Field method, the selection strategy of DDPG has been improved to accelerate the convergence speed during the early stages of training and reduce the time it takes for the mobile robot to reach the target point. Then, optimising the neural network structure of the DDPG algorithm based on the Long Short-Term Memory accelerates the algorithm's convergence speed in complex dynamic scenes. Static and dynamic scene simulation experiments of mobile robots are carried out in ROS. Test findings demonstrate that the Artificial Potential Field method-Long Short Term Memory Deep Deterministic Policy Gradient (APF-LSTM DDPG) algorithm converges significantly faster in complex dynamic scenes. The success rate is improved by 7.3% and 3.6% in contrast to the DDPG and LSTM-DDPG algorithms. Finally, the usefulness of the method provided in this study is similarly demonstrated in real situations using real mobile robot platforms, laying the foundation for the path planning of mobile robots in complex changing conditions.