Zero-shot learning by mutual information estimation and maximization

Shen

The Journal of Engineering

et al. 2022

Zero-shot learning (ZSL) is to identify target categories without labeled data, in which semantic information is used to transfer knowledge from some seen categories. In the existing Generalized Zero-Shot Learning (GZSL) methods, domains shift problem always appeared during generating feature stage. In order to solve this problem, a new method to Auto-Encode the Synthesis Pseudo Features for the GZSL task (AESPF-GZSL) is proposed in this manuscript. Specifically, the AESPF-GZSL method trains the generated features under the semantic auto-encoder framework and exploits attention mechanism to train the generated features again. Then, the generated features are input to the classifier. The proposed method is performed on three benchmark data sets referred as to AWA, CUB and SUN. The experimental results show that the proposed method achieves the state-of-the art classifier accuracy both in ZSL and GZSL settings. In ZSL setting, the classification accuracy of our method is superior to the compared algorithms, improved by 0.40% in AWA and 0.30% in SUN, respectively. And in GZSL setting, the classification accuracy of the method is superior to the comparison algorithm 0.41% in Harmonic mean on AWA, and 1.01%, 0.62%, and 1.05% in training data set, testing data set, and harmonic average on SUN.

Section: Results On Traditional Zslmentioning

confidence: 99%

Auto‐encode the synthesis pseudo features for generalized zero‐shot learning

Shen

The Journal of Engineering

et al. 2022

“…For these different modal data, the DL methods based on multi-modal learning are able to process and learn multi-source and multi-modal information, as well as analyze the states of systems, equipment, environment, and personnel in the nuclear industry from multiple angles [82,83] . Compared with single-modal representation learning where information is expressed as numerical feature vector, the multi-modal learning aligns the features of different modes and eliminates the modes redundancy [84][85][86] .…”

Section: Data Types In Nuclear Industry Chainmentioning

confidence: 99%

“…In addition, DL is difficult to be applied to small-scale, multidimensional, and heterogeneous molecular imaging data. Recently, the proposals of small-sample learning and transfer learning make DL technologies achieve good results in the field of nuclear medicine [85,[88][89][90] . DL methods can be used to realize the image reconstruction, denoising, segmentation, and fusion of molecular imaging.…”

Section: Data Types In Nuclear Industry Chainmentioning

confidence: 99%

Deep learning in nuclear industry: A survey

Tang

Gao

et al. 2022

Big Data Min. Anal.

Self Cite

As a high-tech strategic emerging comprehensive industry, the nuclear industry is committed to the research, production, and processing of nuclear fuel, as well as the development and utilization of nuclear energy.Nowadays, the nuclear industry has made remarkable progress in the application fields of nuclear weapons, nuclear power, nuclear medical treatment, radiation processing, and so on. With the development of artificial intelligence and the proposal of "Industry 4.0", more and more artificial intelligence technologies are introduced into the nuclear industry chain to improve production efficiency, reduce operation cost, improve operation safety, and realize risk avoidance. Meanwhile, deep learning, as an important technology of artificial intelligence, has made amazing progress in theoretical and applied research in the nuclear industry, which vigorously promotes the development of informatization, digitization, and intelligence of the nuclear industry. In this paper, we first simply comb and analyze the intelligent demand scenarios in the whole industrial chain of the nuclear industry. Then, we discuss the data types involved in the nuclear industry chain. After that, we investigate the research status of deep learning in the application fields corresponding to different data types in the nuclear industry. Finally, we discuss the limitation and unique challenges of deep learning in the nuclear industry and the future direction of the intelligent nuclear industry.

Wireless Communications and Mobile Computing

“…Thirdly, the calculation and analysis of the image are used to understand some physical characteristics and related parameters of the specific object in the image. Through the above three data points, it is possible to have a deeper understanding of the specific object in the image and get specific information about the object, but because the computer cannot realize the recognition of the three-dimensional image, it usually has to be converted into a two-dimensional image projection, through one or more two-dimensional image projection to achieve the analysis of the object data [10][11][12][13].…”

Section: Related Workmentioning

confidence: 99%

Deep Learning Distributed Architecture Design Implementation for Computer Vision

Zhang

2022

In the era of big data, to achieve an efficient deep learning and computer vision system for big data, developers need to build a computerized deep learning and computer vision system, and the system can simultaneously complete the tasks of deep learning and computer vision and large-scale data processing. The existing training dataset is reused, and the scene information is small, which cannot meet the needs of large-scale machine training, so it is necessary to include large-scale data, distributed computer system to complete the training. How to meet the training accuracy requirements of deep learning models and minimize the resource cost within the constrained time is a major challenge for distributed deep learning systems. Resource and batch size hyperparameter allocation are the main approaches to optimize the training accuracy and resource cost of models. Existing works have independently configured resources and batch size hyperparameters in terms of computational efficiency and training accuracy, respectively. However, the impact of the two types of configurations on model training accuracy and resource cost has complex dependencies, and it is difficult to achieve the goals of satisfying the model training accuracy requirements and minimizing the resource cost simultaneously by the existing independent configuration methods. To address these problems, this paper proposes a collaborative resource-batch size optimization configuration method for distributed deep learning systems. This method was firstly based on the monotonic function relationship between resource allocation and batch size hyperparameter allocation and model training time and training accuracy, and we select the order-preserving regression theoretical tool to build a model prediction model for single-round complete training time and final training accuracy for computer vision target classification and recognition, respectively; then, we use the abovementioned models together to solve the resource and batch size optimal allocation solutions to meet the model training accuracy requirements with the goal of minimizing resource cost. The optimal allocation of resources and batch size to meet the training accuracy requirements of the model is solved. In this paper, we evaluate the performance of the proposed method for computer vision target recognition based on the proposed distributed deep learning system.