Cognitive Computation最新文献

The primary objective of an effective recommender system is to provide accurate, varied, and personalized recommendations that align with the user’s cognitive intents. Given their ability to represent structural and semantic information effectively, knowledge graphs (KGs) are increasingly being utilized to capture auxiliary information for recommendation systems. This trend is supported by the recent advancements in graph neural network (GNN)-based models for KG-aware recommendations. However, these models often struggle with issues such as insufficient user-item interactions and the misalignment of user intent weights during information propagation. Additionally, they face a popularity bias, which is exacerbated by the disproportionate influence of a small number of highly active users and the limited auxiliary information about items. This bias significantly curtails the effectiveness of the recommendations. To address this issue, we propose a Knowledge-Enhanced User Cognitive Intent Network (KeCAIN), which incorporates item category information to capture user intents with information aggregation and eliminate popularity bias based on causal reasoning in recommendation systems. Experiments on three real-world datasets show that KeCAIN outperforms state-of-the-art baselines.

The remarkable success of Artificial Intelligence in advancing automated decision-making is evident both in academia and industry. Within the plethora of applications, ranking systems hold significant importance in various domains. This paper advocates for the application of a specific form of Explainable AI—namely, contrastive explanations—as particularly well-suited for addressing ranking problems. This approach is especially potent when combined with an Evaluative AI methodology, which conscientiously evaluates both positive and negative aspects influencing a potential ranking. Therefore, the present work introduces Evaluative Item-Contrastive Explanations tailored for ranking systems and illustrates its application and characteristics through an experiment conducted on publicly available data.

Syntactic processing techniques are the foundation of natural language processing (NLP), supporting many downstream NLP tasks. In this paper, we conduct pair-wise multi-task learning (MTL) on syntactic tasks with different granularity, namely Sentence Boundary Detection (SBD), text chunking, and Part-of-Speech (PoS) tagging, so as to investigate the extent to which they complement each other. We propose a novel soft parameter-sharing mechanism to share local and global dependency information that is learned from both target tasks. We also propose a curriculum learning (CL) mechanism to improve MTL with non-parallel labeled data. Using non-parallel labeled data in MTL is a common practice, whereas it has not received enough attention before. For example, our employed PoS tagging data do not have text chunking labels. When learning PoS tagging and text chunking together, the proposed CL mechanism aims to select complementary samples from the two tasks to update the parameters of the MTL model in the same training batch. Such a method yields better performance and learning stability. We conclude that the fine-grained tasks can provide complementary features to coarse-grained ones, while the most coarse-grained task, SBD, provides useful information for the most fine-grained one, PoS tagging. Additionally, the text chunking task achieves state-of-the-art performance when joint learning with PoS tagging. Our analytical experiments also show the effectiveness of the proposed soft parameter-sharing and CL mechanisms.

This article examines the prescribed-time sampled-data control problem for multi-agent systems in signed networks. A time-varying high gain-based protocol is devised to solve the prescribed-time bipartite consensus problem of the linear multi-agent systems with the control gain matrix being resolved through the utilization of the parametric Lyapunov equation. By using the method of scalarization, sufficient conditions are attained to ensure the prescribed-time bipartite consensus of linear multi-agent systems, where the maximum allowable sampling interval (MASI) ensuring the prescribed-time consensus is determined by the initial values of the system state, the linear dynamics of the system, and the maximum eigenvalue of the Laplacian matrix. Specifically, the MASI is inversely proportional to the maximum eigenvalue of the Laplacian matrix. Finally, the validity of the conclusion is ensured through numerical simulation.

In order to obtain a more robust and sparse classifier, in this paper, we propose a novel classifier termed as twin bounded support vector machine with capped pinball loss (CPin-TBSVM), which has the excellent properties of being insensitive to feature and label noise. Given that the proposed model is non-convex, we use the convex-concave procedure algorithm (CCCP) to solve a series of two smaller-sized quadratic programming problems to find the optimal solution. In the process of solving the iterative subproblem, the dual coordinate descent method (DCDM) is used for speeding up solving optimization problems. Moreover, we analyze its theoretical properties, including that the capped pinball loss satisfies Bayes’ rule and CPin-TBSVM has certain noise insensitivity and sparsity. The properties are verified on an artificial dataset as well. The numerical experiment is conducted on 24 UCI datasets and the results are compared with four other models which include SVM, TSVM, Pin-GTSVM and TPin-TSVM. The results show that the proposed CPin-TBSVM has a better classification effect and noise insensitivity.

Over the past few years, larger and deeper neural network models, particularly convolutional neural networks (CNNs), have consistently advanced state-of-the-art performance across various disciplines. Yet, the computational demands of these models have escalated exponentially. Intensive computations hinder not only research inclusiveness and deployment on resource-constrained devices, such as Edge Internet of Things (IoT) devices, but also result in a substantial carbon footprint. Green deep learning has emerged as a research field that emphasizes energy consumption and carbon emissions during model training and inference, aiming to innovate with light and energy-efficient neural networks. Various techniques are available to achieve this goal. Studies show that conventional deep models often contain redundant parameters that do not alter outcomes significantly, underpinning the theoretical basis for model pruning. Consequently, this timely review paper seeks to systematically summarize recent breakthroughs in CNN pruning methods, offering necessary background knowledge for researchers in this interdisciplinary domain. Secondly, we spotlight the challenges of current model pruning methods to inform future avenues of research. Additionally, the survey highlights the pressing need for the development of innovative metrics to effectively balance diverse pruning objectives. Lastly, it investigates pruning techniques oriented towards sophisticated deep learning models, including hybrid feedforward CNNs and long short-term memory (LSTM) recurrent neural networks, a field ripe for exploration within green deep learning research.

The emergence of deep learning (DL) has improved the excellence of generated media. However, with the enlarged level of photorealism, synthetic media is becoming very similar to tangible media, increasing severe problems regarding transmitting fake or deployed data over the Internet. In this situation, it is significant to improve automatic tools to constantly and early identify synthetic media. Generative Adversarial Network (GAN)-based models can create realistic faces that cause deep social and security issues. Existing techniques for identifying GAN-generated faces can execute well on restricted public datasets. Nevertheless, images from existing datasets must signify real situations sufficient for view variants and data distributions, where real faces mainly outnumber artificial ones. Therefore, this study develops an optimal DL-based GAN-generated face detection and classification (ODL-GANFDC) technique. The ODL-GANFDC technique aims to examine the input images properly and recognize whether GAN generates them. To accomplish this, the ODL-GANFDC technique follows the initial stage of the CLAHE-based contrast enhancement process. In addition, the deep residual network (DRN) model must be employed to learn the complex and intrinsic patterns from the preprocessed images. Besides, the hyperparameters of the DRN model can be optimally chosen using an improved sand cat swarm optimization (ISCSO) algorithm. Finally, the GAN-generated faces can be detected using a variational autoencoder (VAE). An extensive set of experimentations can be carried out to highlight the performance of the ODL-GANFDC technique. The experimental outcomes stated the promising results of the ODL-GANFDC technique over compared approaches on the GAN-generated face detection process.

In contemporary real-time applications, diminutive devices are increasingly employing a greater portion of the spectrum to transmit data despite the relatively small size of said data. The demand for big data in cloud storage networks is on the rise, as cognitive networks can enable intelligent decision-making with minimal spectrum utilization. The introduction of cognitive networks has facilitated the provision of a novel channel that enables the allocation of low power resources while minimizing path loss. The proposed method involves the integration of three algorithms to examine the process of big data. Whenever big data applications are examined then distance measurement, decisions mechanism and learning techniques from past data is much importance thus algorithms are chosen according to the requirements of big data and cloud storage networks. Further the effect of integration process is examined with three case studies that considers low resource, path loss and weight functions where optimized outcome is achieved in all defined case studies as compared to existing approach.

In recent years, automatic disease diagnosis has gained immense popularity in research and industry communities. Humans learn a task through both successful and unsuccessful attempts in real life, and physicians are not different. When doctors fail to diagnose disease correctly, they re-assess the extracted symptoms and re-diagnose the patient by inspecting a few more symptoms guided by their previous experience and current context. Motivated by the experience gained from failure assessment, we propose a novel end-to-end automatic disease diagnosis dialogue system called Failure Assessment incorporated Symptom Investigation and Disease Diagnosis (FA-SIDD) Assistant. The proposed FA-SIDD model includes a knowledge-guided, incorrect disease projection-aware failure assessment module that analyzes unsuccessful diagnosis attempts and reinforces the assessment for further investigation and re-diagnosis. We formulate a novel Markov decision process for the proposed failure assessment, incorporating symptom investigation and disease diagnosis frameworks, and optimize the policy using deep reinforcement learning. The proposed model has outperformed several baselines and the existing symptom investigation and diagnosis methods by a significant margin (1–3%) in all evaluation metrics (including human evaluation). The improvements over the multiple datasets and across multiple algorithms firmly establish the efficacy of learning gained from unsuccessful diagnoses. The work is the first attempt that investigate the importance of learning gained from unsuccessful diagnoses. The developed assistant learns diagnosis task more efficiently than traditional assistants and shows robust behavior. Furthermore, the code is available at https://github.com/AbhisekTiwari/FA-SIDA.

With the widespread adoption of deep learning, there has been a notable increase in the prevalence of multimodal deepfake content. These deepfakes pose a substantial risk to both individual privacy and the security of their assets. In response to this pressing issue, researchers have undertaken substantial endeavors in utilizing generative AI and cognitive computation to leverage multimodal data to detect deepfakes. However, the efforts thus far have fallen short of fully exploiting the extensive reservoir of multimodal feature information, which leads to a deficiency in leveraging spatial information across multiple dimensions. In this study, we introduce a framework called Visual-Language Pretraining with Gate Fusion (VLP-GF), designed to identify multimodal deceptive content and enhance the accurate localization of manipulated regions within both images and textual annotations. Specifically, we introduce an adaptive fusion module tailored to integrate local and global information simultaneously. This module captures global context and local details concurrently, thereby improving the performance of image bounding-box grounding within the system. Additionally, to maximize the utilization of semantic information from diverse modalities, we incorporate a gating mechanism to strengthen the interaction of multimodal information further. Through a series of ablation experiments and comprehensive comparisons with state-of-the-art approaches on extensive benchmark datasets, we empirically demonstrate the superior efficacy of VLP-GF.