Neural Computing and Applications最新文献

Alzheimer's disease is a progressive form of dementia. Dementia is a broad term for conditions that impair memory, thinking, and behaviour. Brain traumas or disorders can cause dementia. It is estimated that 60–80% of dementia cases around the world are caused by Alzheimer’s disease, an incurable neurodegenerative disorder. Although Alzheimer's disease research has increased in recent years, early diagnosis is challenging due to the complicated brain structure and functions associated with this disease. It is difficult for doctors to identify Alzheimer's disease in its early stages as there are still no biomarkers to be precise in early detection. In the area of medical imaging, deep learning is becoming increasingly popular and successful. There is no single best approach for the detection of Alzheimer's disease. In comparison with conventional machine learning methods, the deep learning models detect Alzheimer's disease more precisely and effectively. In this review paper, various machine learning-based techniques utilized for the classification of Alzheimer's disease through different imaging modalities are discussed. In addition, a comprehensive and detailed analysis of the various image processing procedures along with corresponding classification performance and feature extraction techniques have been meticulously compiled and presented. The investigation of computer-aided image analysis has demonstrated significant potential in the early detection of cognitive changes in individuals experiencing mild cognitive impairment. Machine learning can provide valuable insights into the cognitive status of patients, enabling healthcare professionals to intervene and provide timely treatment. This review may lead to a reliable method for recognizing and predicting Alzheimer's disease.

Prompt-based portrait image style transfer aims at translating an input content image to a desired style described by text without a style image. In many practical situations, users may not only attend to the entire portrait image but also the local parts (e.g., eyes, lips, and hair). To address such applications, we propose a new framework that enables style transfer on specific regions described by a text description of the desired style. Specifically, we incorporate semantic segmentation to identify the intended area without requiring edit masks from the user while utilizing a pre-trained CLIP-based model for stylizing. Besides, we propose a text-to-patch matching loss by randomly dividing the stylized image into smaller patches to ensure the consistent quality of the result. To comprehensively evaluate the proposed method, we use several metrics, such as FID, SSIM, and PSNR on a dataset consisting of portraits from the CelebAMask-HQ dataset and style descriptions of other related works. Extensive experimental results demonstrate that our framework outperforms other state-of-the-art methods in terms of both stylization quality and inference time.

Methods for socially-aware robot path planning are increasingly needed as robots and humans increasingly coexist in shared industrial spaces. The practice of clearly separated zones for humans and robots in shop floors is transitioning towards spaces where both humans and robot operate, often collaboratively. To allow for safer and more efficient manufacturing operations in shared workspaces, mobile robot fleet path planning needs to predict human movement. Accounting for the spatiotemporal nature of the problem, the present work introduces a spatiotemporal graph neural network approach that uses graph convolution and gated recurrent units, together with an attention mechanism to capture the spatial and temporal dependencies in the data and predict human occupancy based on past observations. The obtained results indicate that the graph network-based approach is suitable for short-term predictions but the rising uncertainty beyond short-term would limit its applicability. Furthermore, the addition of learnable edge weights, a feature exclusive to graph neural networks, enhances the predictive capabilities of the model. Adding workspace context-specific embeddings to graph nodes has additionally been explored, bringing modest performance improvements. Further research is needed to extend the predictive capabilities beyond the range of scenarios captured through the original training, and towards establishing standardised benchmarks for testing human motion prediction in industrial environments.

Conventional authentication methods, such as passwords and PINs, are vulnerable to multiple threats, from sophisticated hacking attempts to the inherent weaknesses of human memory. This highlights a critical need for a more secure, convenient, and user-friendly approach to authentication. This paper introduces M2auth, a novel multimodal behavioral biometric authentication framework for smartphones. M2auth leverages a combination of multiple authentication modalities, including touch gestures, keystrokes, and accelerometer data, with a focus on capturing high-quality, intervention-free data. To validate the efficacy of M2auth, we conducted a large-scale field study involving 52 participants over two months, collecting data from touch gestures, keystrokes, and smartphone sensors. The resulting dataset, comprising over 5.5 million action points, serves as a valuable resource for behavioral biometric research. Our evaluation involved two fusion scenarios, feature-level fusion and decision-level fusion, that play a pivotal role in elevating authentication performance. These fusion approaches effectively mitigate challenges associated with noise and variability in behavioral data, enhancing the robustness of the system. We found that the decision-level fusion outperforms the feature level, reaching a 99.98% authentication success rate and an EER reduced to 0.84%, highlighting the robustness of M2auth in real-world scenarios.

Personalized recommendation plays a crucial role in Internet platforms, providing users with tailored content based on their user models and enhancing user satisfaction and experience. To address the challenge of information overload, it is essential to analyze user needs comprehensively, considering historical behavior and interests and the user's intentions and profiles. Previous user modeling approaches for personalized recommendations have exhibited certain limitations, relying primarily on historical behavior data to infer user preferences, which results in challenges such as the cold-start problem, incomplete modeling, and limited explanation. Motivated by recent advancements in large language models (LLMs), we present a novel approach to user modeling by embracing generative user modeling using LLMs. We propose generative user modeling with chain-of-thought prompting for personalized recommendation, which utilizes LLMs to generate comprehensive and accurate user models expressed in natural language and then employs these user models to empower LLMs for personalized recommendation. Specifically, we adopt the chain-of-thought prompting method to reason about user attributes, subjective preferences, and intentions, integrating them into a holistic user model. Subsequently, we utilize the generated user models as input to LLMs and design a collection of prompts to align the LLMs with various recommendation tasks, encompassing rating prediction, sequential recommendation, direct recommendation, and explanation generation. Extensive experiments conducted on real-world datasets demonstrate the immense potential of large language models in generating natural language user models, and the adoption of generative user modeling significantly enhances the performance of LLMs across the four recommendation tasks. Our code and dataset can be found at https://github.com/kwyyangfan/GUMRec.

Double-layer coating of dextran and citrate (DC) on the Fe₃O₄ (magnetite) ferrofluids (FFs) has been conducted for biomedical applications such as hyperthermia and magnetic resonance imaging. The magnetic retardance of DC-coated FFs was measured, and the magnetic heating effect was investigated previously. An experiment was conducted using the uniform design method; we enabled the formula to fit with experimental data on retardance through the stepwise regression analysis. Two intelligent search methods, particle swarm optimization (PSO) and simulated annealing (SA), were used to find the maximum retardance. The optimized parametric combinations were decided as [0.0750, 75.7945, 0.3225, 0.6500] and [0.0750, 75.844, 0.323, 0.65], respectively, denoting the Fe₃O₄ concentration, the coating temperature, the mass of citrate and dextran. The corresponding maximum retardances were 119.6576° and 119.6558°. The PSO algorithm was more effective and accessible than the SA algorithm in optimizing retardance. As for the hybrid optimization selected for solving complex problems, such as PSO was used to find a near-global solution, and SA was then used for searching for a global solution, the parameter fine-tuning of SA affects the final result. A hybrid metaheuristic algorithm with the local gradient-based sequential quadratic programming (SQP) algorithm is used to find the global solution because of its effectiveness and convergence speed in research. Overall, we provide some two-level hybrid optimizations for the global exploration of the retardance of DC-coated FFs. The hybrid algorithms, including PSO-SA, PSO-SQP, or SA-SQP, allow us to explore a more accurate global solution with high performance.

The advances in machine learning (ML) tools and techniques have enabled the non-intrusive collection and rapid analysis of massive amounts of data involving athletes in competitive collegiate sports. It has facilitated the development of services that a coach can employ in analyzing these data into actionable insights in designing training schedules and effective strategies for maximizing an athlete’s performance, while minimizing injury risk. Collegiate sports utilize data to get a competitive advantage. While game statistics are publicly available, relying on more than one form of data can help reveal a pattern. We developed a framework that considers various modalities and creates an athletic signature to predict their future performance. Our research involves the study of 42 distinct features that quantify various internal/external stressors the athletes face to characterize and estimate their athletic readiness (in the form of reactive strength index modified—RSImod) using ML algorithms. Our study, conducted over 26 weeks with 17 collegiate women’s basketball athletes, developed a framework that first performed sensitivity analysis using a hybrid approach combining the strengths of various filter-based, wrapper-based, and embedded feature importance techniques to identify the features most significantly impacting athlete readiness. These features were then categorized into four moderating variables (MVs, i.e. factors): sleep, cardiac rhythm, training strain, and travel schedule. Further, we used factor analysis to enhance interpretability and reduce computational complexity. A hybrid boosted-decision-trees-based model designed based on athlete clusters predicted future athletic readiness based on MVs with a mean squared error (MSE) of 0.0102. Partial dependence plots (PDPs) helped qualitatively learn the relationship between the moderating variables and the RSImod score. Athletic signatures, uniquely defining athlete-specific MV patterns, account for intra-individual variability, offering a better statistical basis for predicting game lineup (green/yellow/red card assignment) in combination with model predictions. SHAP (SHapley Additive exPlanations) values suggest the causative MV in order of significance for each prediction, enabling coaches to make informed decisions about training adjustments and athlete well-being, leading to performance improvement. Using the fingerprint mechanism, we created green (within 1 Standard Deviation (SD)), yellow (> 1SD and < 2SD), and red card (> 2SD) zones for athlete readiness assessment. While, this study was conducted on D-I women’s basketball, the modalities apply to several sports, such as soccer, volleyball, football, and ice hockey. This framework allows coaches to understand their athlete dynamics from a strictly data perspective, which helps them strategize their next moves, combined with their personal experience and interactions with the team.

Additive manufacturing, particularly laser powder bed fusion (L-PBF), is an emerging method for fabricating complex parts in various industries. However, it faces the persistent challenge of thermal deformation, a significant barrier to its wider application and reliability. Current strategies, while partially effective, do not fully address the intricate thermal dynamics of the process, indicating a clear research gap in optimizing manufacturing techniques for better thermal management. This study focuses on understanding and mitigating thermal deformation in L-PBF using Genetic Algorithms (GAs). The application of GAs as a ‘black-box’ approach is explored to gain insights into the complex physics of L-PBF. A comprehensive investigation into the optimization of island sequencing within L-PBF processes is presented, employing GAs to systematically reduce thermal deformation. Various island sequences in a bilayered block structure are analyzed to assess the effectiveness of GAs in minimizing deformation, including scenarios such as variations in block sizes and interlayer rotation angles. Statistical tools such as silhouette scores and probability density distribution plots are utilized to provide a thorough analysis of deformation patterns and their respective thermal behaviors. The results show GA's remarkable efficiency in enhancing thermal management, achieving a significant reduction in thermal deformation within a range of 12–15% across the examined scenarios. This achievement highlights GA's capability in rapid optimization of scan sequences for better thermal deformation control. The findings enhance the understanding of thermal dynamics in L-PBF and consequently open new avenues for improving the quality and reliability of other metal additive manufacturing processes as well.

Wireless Sensor Networks (WSNs) are essential for collecting and transmitting data in modern applications that rely on data, where effective network connectivity and coverage are crucial. The optimal placement of router nodes within WSNs is a fundamental challenge that significantly impacts network performance and reliability. Researchers have explored various approaches using metaheuristic algorithms to address these challenges and optimize WSN performance. This paper introduces a new hybrid algorithm, CFL-PSO, based on combining an enhanced Fick’s Law algorithm with comprehensive learning and Particle Swarm Optimization (PSO). CFL-PSO exploits the strengths of these techniques to strike a balance between network connectivity and coverage, ultimately enhancing the overall performance of WSNs. We evaluate the performance of CFL-PSO by benchmarking it against nine established algorithms, including the conventional Fick’s law algorithm (FLA), Sine Cosine Algorithm (SCA), Multi-Verse Optimizer (MVO), Salp Swarm Optimization (SSO), War Strategy Optimization (WSO), Harris Hawk Optimization (HHO), African Vultures Optimization Algorithm (AVOA), Capuchin Search Algorithm (CapSA), Tunicate Swarm Algorithm (TSA), and PSO. The algorithm’s performance is extensively evaluated using 23 benchmark functions to assess its effectiveness in handling various optimization scenarios. Additionally, its performance on WSN router node placement is compared against the other methods, demonstrating its competitiveness in achieving optimal solutions. These analyses reveal that CFL-PSO outperforms the other algorithms in terms of network connectivity, client coverage, and convergence speed. To further validate CFL-PSO’s effectiveness, experimental studies were conducted using different numbers of clients, routers, deployment areas, and transmission ranges. The findings affirm the effectiveness of CFL-PSO as it consistently delivers favorable optimization results when compared to existing methods, highlighting its potential for enhancing WMN performance. Specifically, CFL-PSO achieves up to a 66.5% improvement in network connectivity, a 16.56% improvement in coverage, and a 21.4% improvement in the objective function value when compared to the standard FLA.

The objective of this research is to develop some novel operational laws based of T-norm and T-conorm and then using these operational laws to develop several Einstein operators for aggregating the different complex intuitionistic fuzzy numbers (CIFNs) by considering the dependency between the pairs of its membership degrees. In the existing studies of fuzzy and its extensions, the uncertainties present in the data are handled with the help of degrees of membership that are the subset of real numbers, which may also loss some valuable data and hence consequently affect the decision results. A modification to these, complex intuitionistic fuzzy set handles the uncertainties with the degree whose ranges are extended from real subset to the complex subset with unit disk and hence handle the two-dimensional information in a single set. Thus, motivated by this and this paper we present some novel methods such as complex intuitionistic fuzzy Einstein weighted geometric aggregation (CIFEWGA) operator, complex intuitionistic fuzzy Einstein ordered weighted geometric aggregation (CIFEOWGA) operator, complex intuitionistic fuzzy Einstein hybrid geometric aggregation (CIFEHGA) operator, induced complex intuitionistic fuzzy Einstein ordered weighted geometric aggregation (I-CIFEOWGA) operator and induced complex intuitionistic fuzzy Einstein hybrid geometric aggregation (I-CIFEHGA) operator. We present some of their desirable properties such as idempotency, boundedness and monotonicity. Furthermore, based on these methods a multi-attribute group decision-making problem developed under complex intuitionistic fuzzy set environment. An illustrative example related to the selection of the best alternative is considered to show the effectiveness, importance and efficiency of the novel approach.