Railroad travel has become an essential part of modern life, therefore, rail surface defect detection has become a problem that cannot be ignored. However, most of the currently existing methods often require a large ...
详细信息
This paper selects the evaluation indexes that fit with the teaching mode and evaluation objectives of university innovation and entrepreneurship education in the new media environment. The limitations of the existing...
详细信息
Take-over performance plays a significant role in evaluating drivers' state, and serves as a crucial reference for enhancing control transitions in the context of conditionally automated driving. In this study, we...
详细信息
Take-over performance plays a significant role in evaluating drivers' state, and serves as a crucial reference for enhancing control transitions in the context of conditionally automated driving. In this study, we aim to predict minimum anticipated collision time (min ACT), an indicator of drivers' take-over performance, in expectation of promoting safer take-overs via deep learning, so that drivers' state detriment of take-over safety could be adjusted accordingly with intelligent human-machine interaction algorithms predictably. By incorporating multi-source information including drivers' state, drivers' demographics, surrounding traffic features as well as driver-vehicle interaction characteristics, network model “ACTNet” was proposed to facilitate continuous estimation. Depthwise separable convolution and non-local self-attention were utilized to prevent overfitting and establish spatial dependency over fixation heatmap, respectively. To overcome data distribution imbalance, class balanced loss was used in conjunction with regression loss to realize more accurate predictions. Driving simulator experiment was conducted with dataset collected for the subsequent verification of the proposed algorithm. Potentialities of deep learning methods were highlighted for take-over studies, contributing to the design of intelligent human-machine interaction systems in conditional automation. Our findings present a valid method of deep learning in predicting drivers' take-over performance and meanwhile have implications for the development of intelligent adaptive take-over time budget regulation and dynamic drivers' state adjustment algorithms. IEEE
Reinforcement learning has been successfully applied in various fields, such as games and robots. However, there are still some issues in the traditional reinforcement learning paradigm that involves one agent per env...
Reinforcement learning has been successfully applied in various fields, such as games and robots. However, there are still some issues in the traditional reinforcement learning paradigm that involves one agent per environment, such low efficiency in data generation and insufficient amount of training data. In this paper, we propose a federated reinforcement learning framework that utilizes multiple agents to generate data for simultaneous training of models. Specifically, the proposed framework allows the best performing agent to share its learning experience with other agents for improving the learning performance and protecting the privacy of agents. The proposed framework is applied to the Cart-pole, Acrobot and LunarLander environments in OpenAI Gym, and the results show that a significant improvement of the learning performance can be achieved by adopting our framework.
In this paper, we focus on the task of image retrieval with text feedback, which maintains two key challenges. One is the misalignment problem between different modalities, and the other is to selectively alter the co...
In this paper, we focus on the task of image retrieval with text feedback, which maintains two key challenges. One is the misalignment problem between different modalities, and the other is to selectively alter the corresponding attributes on the reference image according to the textual words. To this end, we propose a novel visual-linguistic alignment and composition network (ACNet) consisting of two key components: the modality alignment module (MAM) and the relation composition module (RCM). Specifically, the MAM performs alignment between the features from different modalities by applying image-text contrastive loss. The RCM correlates the image regions with their corresponding words and then adaptively modifies the specific regions of the reference image conditioned on textual semantics. Quantitative and Qualitative experiments on three datasets not only demonstrate that our ACNet outperforms state-of-the-art models, but also verify the effectiveness of our method.
With the development of deep learning on EEG-related tasks, the complexity of learning models has gradually increased. Unfortunately, the insufficient amount of EEG data limits the performance of complex models. Thus,...
With the development of deep learning on EEG-related tasks, the complexity of learning models has gradually increased. Unfortunately, the insufficient amount of EEG data limits the performance of complex models. Thus, model compression becomes an option to be seriously considered. So far, in EEG- related tasks, although some models used lightweight means such as separable convolution in their models, no existing work has directly attempted to compress the EEG model. In this paper, we try to investigate the state-of-the-art network pruning methods on commonly used EEG models for the emotion recognition task. In this work, we make several surprising observations that contradict common beliefs. Training a pruned model from scratch outperforms fine-tuning a pruned model with inherited weights, which means that the pruned structure itself is more important than the inherited weights. We can ignore the entire pruning pipeline and train the network from scratch using the predefined network architecture. We substantially reduce the computational resource overhead of the model while maintaining accuracy. In the best case, we achieve a 62.3% reduction in model size and a 64.3% reduction in computing operations without accuracy loss.
Anomaly detection in dynamic graphs has drawn increasing attention in social networks, e-commerce, and cybersecurity etc. Capturing the evolution patterns using temporal features is crucial for dynamic graphs anomaly ...
Anomaly detection in dynamic graphs has drawn increasing attention in social networks, e-commerce, and cybersecurity etc. Capturing the evolution patterns using temporal features is crucial for dynamic graphs anomaly detection. However, existing approaches do not thoroughly consider the long-term and short-term temporal features, which lead to the loss of information about the evolution patterns of dynamic graphs. In this paper, we propose a end-to-end an anomaly detection algorithm in dynamic graphs via Long Short-term Temporal Attention network (GraphLSTA). GraphLSTA captures the evolution patterns of dynamic graphs by effectively extracting and fusing the long-term and short-term temporal features through a recurrent attention mechanism, improving the detection performance.
In this paper, we propose a novel approach to enhance user and restaurant representations in the context of predicting the closure of a restaurant and give an explanation based on data generated from user-restaurant i...
In this paper, we propose a novel approach to enhance user and restaurant representations in the context of predicting the closure of a restaurant and give an explanation based on data generated from user-restaurant interactions. In order to accurately predict the operating status of a restaurant and give a reasonable explanation, we need rich relevant information of user-restaurant interaction and reviews to model their representations. However, the interaction information between the user and the restaurant is usually sparse. To address this issue, we propose a new model, which is called the Co-Attentive Contrastive Learning (CACL) model. Our model employs a contrastive learning algorithm to deal with data sparsity and leverages co-attention mechanism to select the most relevant review information assisting the former to obtain a more accurate and granular representation. By fusing these module, we obtain rich information to perform two tasks with better quality. To demonstrate the effectiveness of our model, extensive experiments was conducted in six different cities, and the results showed that CACL was superior to the previous method in terms of prediction accuracy and explainable ability (average 4.8% improvement in prediction and 35.4% improvement in explanation).
Dynamic constrained multi-objective optimization problems (DCMOPs) are characterized by time-varying objectives and constraints, requiring optimization algorithms that can rapidly track the changing Pareto-Optimal Set...
Dynamic constrained multi-objective optimization problems (DCMOPs) are characterized by time-varying objectives and constraints, requiring optimization algorithms that can rapidly track the changing Pareto-Optimal Set (POS).A new dynamic constrained multi-objective evolutionary algorithm with adaptive two-stage archiving and autoencoder prediction is proposed in this paper, called ATAP to effectively solve DCMOPs. Specifically, ATAP designs a differential denoising autoencoder (DDA) prediction strategy, which applies a denoising autoencoder to thoroughly analyze change trends of historical population and predict some initial solutions in the new environment. Subsequently, to promote greater diversity within the initial population, a differential rule is implemented, effectively addressing the potential scarcity of diversity caused by constraints. Moreover, ATAP introduces an adaptive two-stage archiving (ATA) constraint handling technique, which can dynamically adjust evolutionary stages based on the state of the population. This approach can adaptively determine preferences between objectives and constraints, achieving a better balance. In this way, ATA and DDA are well cooperated to efficiently solve DCMOPs with time-varying constraints and objectives. The experimental results demonstrate that the proposed ATAP is effective and has some advantages over three competitive algorithms when solving the CEC2023 DCMOP benchmark.
Bias is a common problem in both human cognition and machine learning tasks. However, machines struggle more than humans with bias reduction, mainly because most algorithms rely on the assumption that the training dat...
Bias is a common problem in both human cognition and machine learning tasks. However, machines struggle more than humans with bias reduction, mainly because most algorithms rely on the assumption that the training data represent a full population in the real world. This assumption does not hold in most cases, i.e., collecting unbiased or representative data is challenging due to various factors such as sampling errors, human prejudices, and cultural differences. Therefore, bias mitigation has become an essential task in artifical intelligence. In recent years, significant progress has been made in improving the effectivess of debiasing, especially in computer vision. This paper provides a comprehensive review of de-biasing methods for neural networks trained on image data. Firstly, we present a formal defination of the bias mitigation problem and discuss some relevant topics. Moreover, we categorize the existing methods into three main types, namely, data-level methods that aim to balance or augment the training data, model-level methods that aim to modify or regularize the learning algorithms, and adversarial and ensembling approaches that use additional models to capture the biases present in data and to guide the training of the primary model. Finally, we conclude this survey and suggest potential research directions for the future.
暂无评论