Knowledge-Based Systems最新文献

Learning from positive and unlabeled data (PU learning) is actively researched machine learning task. The goal is to train a binary classification model based on a training dataset containing part of positives which are labeled, and unlabeled instances. Unlabeled set includes remaining part of positives and all negative observations. An important element in PU learning is modeling of the labeling mechanism, i.e. labels’ assignment to positive observations. Unlike in many prior works, we consider a realistic setting for which probability of label assignment, i.e. propensity score, is instance-dependent. In our approach we investigate minimizer of an empirical counterpart of a joint risk which depends on both posterior probability of inclusion in a positive class as well as on a propensity score. The non-convex empirical risk is alternately optimized with respect to parameters of both functions. In the theoretical analysis we establish risk consistency of the minimizers using recently derived methods from the theory of empirical processes. Besides, the important development here is a proposed novel implementation of an optimization algorithm, for which sequential approximation of a set of positive observations among unlabeled ones is crucial. This relies on modified technique of ’spies’ as well as on a thresholding rule based on conditional probabilities. Experiments conducted on 20 data sets for various labeling scenarios show that the proposed method works on par or more effectively than state-of-the-art methods based on propensity function estimation.

In federated learning (FL), ensuring the efficiency of global models generated from the weighted aggregation of local models with data heterogeneity remains challenging. Moreover, the contradiction between imprecise aggregation weights and changing data distributions leads to aggregation errors that increase in an accelerated manner throughout the process. Therefore, we present federated learning using adaptive aggregate weights (FedAAW) to change the optimization direction in steps, including local training and global aggregation, and reduce inefficiencies in the global model due to the accelerated growth of aggregation errors resulting from changes in heterogeneity. In each round, the global- and local-model information is dynamically combined to generate an initial model at the beginning of the local training. The key module in FedAAW is adaptive aggregate weights (AAW), which are used to update the aggregation weight by sharing an optimization objective with global training and using the gradient information from other clients to accurately compute the updated aggregation weight direction. AAW guarantee consistency between weight update and global optimization, theoretically demonstrating convergence. The results of our comprehensive experiments on public datasets demonstrate that the test accuracy metrics of FedAAW are higher than those of six state-of-the-art algorithms and that FedAAW is capable of up to 50% improvement. FedAAW also results in an improvement of 14% on CIFAR100, a complex dataset, when compared with the best-performing baseline. FedAAW is faster than other algorithms in attaining the specified accuracy in experiments; in particular, it is approximately three times faster than federated learning with adaptive local aggregation. In addition, the results obtained in experimental environments with different client sizes and heterogeneous data confirm that FedAAW is robust and scalable.

In this paper, we introduce the Semantic Environment Atlas (SEA), a novel mapping approach designed to enhance visual navigation capabilities of embodied agents. The SEA utilizes semantic graph maps that intricately delineate the relationships between places and objects, thereby enriching the navigational context. These maps are constructed from image observations and capture visual landmarks as sparsely encoded nodes within the environment. The SEA integrates multiple semantic maps from various environments, retaining a memory of place-object relationships, which proves invaluable for tasks such as visual localization and navigation. We developed navigation frameworks that effectively leverage the SEA, and we evaluated these frameworks through visual localization and object-goal navigation tasks. Our SEA-based localization framework significantly outperforms existing methods, accurately identifying locations from single query images. Experimental results in Habitat Savva et al. (2019)scenarios show that our method not only achieves a success rate of 39.0%—an improvement of 12.4% over the current state-of-the-art—but also maintains robustness under noisy odometry and actuation conditions, all while keeping computational costs low.

Multi-agent value decomposition (MAVD) algorithms have made remarkable achievements in applications of multi-agent reinforcement learning (MARL). However, overestimation errors in MAVD algorithms generally lead to unstable phenomena such as severe oscillation and performance degradation in their learning processes. In this work, we propose a method to integrate the advantages of value averaging and operator switching (VAOS) to enhance MAVD algorithms’ learning stability. In particular, we reduce the variance of the target approximate error by averaging the estimate values of the target network. Meanwhile, we design a operator switching method to fully combine the optimal policy learning ability of the Max operator and the superior stability of the Mellowmax operator. Moreover, we theoretically prove the performance of VAOS in reducing the overestimation error. Exhaustive experimental results show that (1) Comparing to the current popular value decomposition algorithms such as QMIX, VAOS can markedly enhance the learning stability; and (2) The performance of VAOS is superior to other advanced algorithms such as regularized softmax (RES) algorithm in reducing overestimation error.

Lip-reading classification has received a lot of interest in recent decades because it is widely used in a variety of fields. It plays an important role in interpreting spoken words in noisy situations and reconstructing communication processes for those with hearing impairments. Despite significant advancements in this field, there are still several drawbacks in existing work such as feature extraction and Model capability for visual speech recognition. For these reasons, the current paper suggests an Optimized Quaternion Meixner Moments Convolutional Neural Network (OQMMs-CNN) method that intends to develop a Visual Speech Recognition (VSR) system based only on video images. This unique method combines OQMMs optimized for the GWO algorithm and convolutional neural networks taken from deep learning techniques with the aim of recognizing digits, words, or letters displayed as input videos.The OQMMs are used here as descriptors with the purpose of identifying, holding, and extracting essential information from video images (lips images) and generating moments for CNN input. The latter uses Meixner polynomials, which are defined by local parameters α and β. Then, the Grey Wolf optimization method (GWO) is applied to enssure excellent classification accuracy by optimizing those local parameters. After being tested on three public datasets such as AVLetters, Grid, AVDigits, and LRW, and comparing to several ways using complicated models and deep architecture, the method emerges as an excellent solution for reducing the high dimensionality of video pictures and training time.

Knowledge Graph Embedding (KGE) based deep neural networks contribute to recommender systems in diverse application scenarios. However, Catastrophic Forgetting (CForg) significantly degrades their performance. Although exemplar replay is commonly adopted as a possible remedy to alleviate the intensity of CForg, a trade-off between performance and complexity occurs in this process. Therefore, in this work, we introduce Continual Knowledge graph embedding enhancement for joint Interaction-based Next click recommendation (CKIN) to defy the CForg and assuage the complexity. Typically, we introduce the Semantic Relevance Estimation (SRE) technique to ensure information relevance by filtering out irrelevant-data and reducing the space complexity. We introduce the SRE-enhanced deep probabilistic technique to probably replay the most relevant exemplars to defy the CForg and reduce the time complexity. Moreover, we introduce the integration of locality-preserving loss into the KGE framework to optimize the loss. In substantial experiments on real-world datasets, CKIN outperforms the baseline methods by effectively meeting the highlighted challenges.

The accurate prediction of solar energy generation is significant for efficient energy management in Internet of Things (IoT) devices. However, current forecasting models frequently fail to account for the dynamic nature of weather conditions. Also, traditional methods often have limited accuracy and scalability. This paper proposes Solar Energy Prediction in IoT system based optimized Complex-Valued Spatio-Temporal Graph Convolutional Neural Network (SEP-CVSGCNN-IoT) to overcome the limitations of existing models. Initially, the data are collected from solar panel and weather forecast. The collected data is given to the pre-processing using Data-Adaptive Gaussian Average Filtering (DAGAF) to remove the unwanted data and replace missing data. The pre-processed data is given into Nutcracker Optimization (NCO) algorithm for selecting optimal features. Then, the selected features are given to the Complex-Valued Spatio-Temporal Graph Convolutional Neural Network (CVSGCNN) for solar energy prediction. Finally, Dipper Throated Optimization Algorithm (DTOA) is proposed to enhance the weight parameter of CVSGCNN classifier, which precisely predicts solar energy in IoT. The proposed SEP-CVSGCNN-IoT method provides 18.46%, 26.34, 15.69 and 20.84% higher accuracy and 18.24%, 23.77, 24.34 and 16.29% lower mean absolute error when analyzed with existing techniques, such as deep learning enhanced solar energy prediction and AI-driven IoT (DL-ESEF-AI), towards efficient renewable energy prediction using deep learning (TEE-REP-DL), a new deep learning method for effectual forecasting of short-term PV energy production (DL-EF-SPEP) and metaheuristic-dependent hyper parameter tuning for recurrent deep learning: application to the solar energy generation prediction (HT-RDL-PSEG) respectively.

Unsupervised Domain Adaptation (UDA) is a promising strategy for representing unlabeled data through domain alignment. Nonetheless, a considerable number of whole-domain alignment techniques often neglect the essential interconnections between pixels and patches across distinct domains that exhibit analogous semantic characteristics. This oversight can hinder their ability to manage semantic variations across domains and to create a discriminative embedding for different classes, ultimately leading to reduced discrimination and poor generalization. This paper presents a novel UDA method for medical image analysis, termed CMRVAE. The proposed method is composed of a margin-restrained variational auto-encoder (MR-VAE) and a class-separation patch-level manifold clustering (CPMC) module. The MR-VAE embeds an adaptive margin-based enhancement technique through an innovative variational inference for optimal encoder mapping in UDA. The CPMC module integrates multi-granularity class information into the manifold for improved preparatory work before UDA. Experimental results on three cardiac datasets show that the proposed method achieves substantially enhanced accuracy compared to the state-of-the-art unsupervised approaches.

The massive spread of rumors on social networks has caused serious adverse effects on individuals and society, increasing the urgency of rumor detection. Existing detection methods based on deep learning have achieved fruitful results by virtue of their powerful semantic representation capabilities. However, the centralized training mode and the reliance on extensive training data containing user privacy pose significant risks of privacy abuse or leakage. Although federated learning with client-level differential privacy provides a potential solution, it results in a dramatic decline in model performance. To address these issues, we propose a Federated Graph Knowledge Distillation framework (FedGKD), which aims to effectively identify rumors while preserving user privacy. In this framework, we implement anonymization from both the feature and structure dimensions of graphs, and apply differential privacy only to sensitive features to prevent significant deviation in data statistics. Additionally, to improve model generalization performance in federated settings, we learn a lightweight generator at the server to extract global knowledge through knowledge distillation. This knowledge is then broadcast to clients as inductive experience to regulate their local training. Extensive experiments on four publicly available datasets demonstrate that FedGKD outperforms strong baselines and displays outstanding privacy-preserving capabilities.