China 
Africa 
Explore chapters and articles related to this topic
Cancer Diagnosis from Histopathology Images Using Deep Learning: A Review
Published in Ranjeet Kumar Rout, Saiyed Umer, Sabha Sheikh, Amrit Lal Sangal, Artificial Intelligence Technologies for Computational Biology, 2023
Vijaya Gajanan Buddhavarapu, J. Angel Arul Jothi
Generative adversarial networks (GANs) are deep generative models that consist of two main structures: generator and discriminator [26]. The discriminator is a classifier network. The generator network transforms the input sample vector into a possibly meaningful output. During training, the following steps take place: The generator takes in input a sample vector (usually random noise) and outputs transformed sample. When the model is image-based, this transformed sample is a fake image.The transformed sample is then fed into the discriminator network that is tasked with distinguishing whether the sample is real or fake.Using the discriminator output, the weights of discriminator and generator networks are updated.The discriminator is penalized when fake data is predicted as real data and the generator is penalized when fake data is predicted as fake data.
Background and Related Methods
Published in Rui Yang, Maiying Zhong, Machine Learning-Based Fault Diagnosis for Industrial Engineering Systems, 2022
The generative adversarial networks are based on machine learning and game theory, and the main idea is to recognize the generated data from the original data until the generated data is recognized as real data. The generative adversarial network is a generative model based on adversarial ideas, consisting of two parts: generator (G) and discriminator (D). Generative adversarial network completes the task through the generator to generate data that can be identified as “true” by the discriminator through multiple iterations of training. First, the generator generates data through random noise, and then inputs these data into the discriminator. Then the discriminator uses the real input data as the basis for identification of the input generated data. If the identification is “true”, it indicates that the generated data has been considering similar to the real data. If the identification is “false”, the discriminator will feed back the result to the generator and instruct the generator to generate data again. The idea of generative and adversary has been applied to the field of transfer learning in recent years to reduce the difference between the feature distributions of two domains.
Adaptive Real-Time Underwater Visual Restoration with Adversarial Critical Learning
Published in Junzhi Yu, Xingyu Chen, Shihan Kong, Visual Perception and Control of Underwater Robots, 2021
Junzhi Yu, Xingyu Chen, Shihan Kong
Recently, Generative Adversarial Networks (GAN) [11] have been successfully employed in image-to-image translation tasks, e.g., style transfers and super-resolution [12]. It is clear that image restoration can be treated as an image-to-image translation, so we are certain that GAN is able to restore the underwater scenes if trained with paired data (i.e., original underwater images and corresponding in-air versions). Furthermore, a well-trained GAN-based method can adaptively work for various underwater scenarios. When it comes to underwater training data, although paired images are hard to be obtained, synthetic in-air data based on a traditional method can provide unambiguous visual content for training. However, the characteristics of synthetic samples and real in-air data are still distinct to some extent, so synthetic images cannot be employed as the ground truth. Otherwise, GAN’s results can perform similarly but no better than the synthetic data. That is, underwater noise that incurs color distortion, contrast decrease, and haziness still needs to be further removed. Thereby, a new framework is required for further enhancement.
Transfer-GAN: data augmentation using a fine-tuned GAN for sperm morphology classification
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2023
Amir Abbasi, Sepideh Bahrami, Tahere Hemmati, Seyed Abolghasem Mirroshandel
Considering the outcomes of our method, now it is good to discuss the use of our approach in the real world. A real-world sperm classifier system should be trained with a wide range of samples. The MHSMA and HuSHem datasets are small and previous successful works were grappling with data limitation and attempted to tackle the problem using transfer learning. In this study, we focused on proposing the method which is able to solve the data limitation problem directly by extending the dataset. Collecting more data is a very time-consuming and expensive task. In addition, high-tech microscopes and lots of experts are also needed. To deal with this problem, usually classical data augmentation techniques are employed to extend datasets using modified copies of images. Nevertheless, these techniques can produce a limited number of unseen samples. GAN models basically learn distribution of the training images and generate similar images which leads to having new different samples of our dataset. By considering the performance enhancement of used benchmark models, at least our method can be used to solve their data limitation problem and make the proposed techniques more reliable to be used in fertility clinics. In addition, our outcomes confirm that synthetic images are reliable and have the specific features which is hidden within real images. Hence, in addition to deep transfer based classifiers, it can be suggested that using our method in the process of training other existing sperm abnormality systems can improve their performance and solve data limitation.
Highlight Removal from A Single Grayscale Image Using Attentive GAN
Published in Applied Artificial Intelligence, 2022
Haitao Xu, Qiang Li, Jing Chen
Since the generative adversarial network (Goodfellow et al. 2014) was proposed in 2014, it has achieved great success in the fields of deep learning (Gui et al. 2020). There are many applications of GAN in computer vision, such as image inpainting (Yeh et al. 2016; Yu et al. 2018), super-resolution (Ding et al. 2019; Xintao Wang et al. 2018b), and image translation (Isola et al. 2017; Zhu et al. 2017). Inspired by the success of GAN in image translation, we utilize it to implement the specular highlight removal. Some scholars have made related attempts before. John Lin et al. used a multi-class discriminator to train the generator to remove the specular highlights from color images (Lin et al. 2019). Funke et al. employed GAN in the endoscope highlights removal(Funke et al. 2018). In general, the research on the combination of GAN and specular highlight removal is still at the initial stage. As discussed in the introduction, this research topic is of great research value, so we propose a GAN-based method to remove specular highlights from a single grayscale image in this paper.
Deep Learning Techniques for OFDM Systems
Published in IETE Journal of Research, 2021
M. Meenalakshmi, Saurabh Chaturvedi, Vivek K. Dwivedi
The generative adversarial networks contain two NNs called generator and discriminator in which the generator NN generates data instances, and discriminator NN evaluates the generated data instances by comparing it with true data instances. All NNs with multiple hidden layers are called DNNs. The learning capability of a DNN is very high, and it effectively learns the complex functions. The spiking NN accurately mimics the functionality of a biological NN in which the neurons use spikes to communicate with each other, and the neurons are called spiking neurons. In the feedforward NN, the data flow from the input layer to the output layer in one direction without any loops. Depending upon the architecture, the feedforward NNs are classified as convolutional NN [24], extreme learning machines, time delay NN, autoencoder, radial basis function, and probabilistic NN. In the physical NN, the activation of a neuron is done by adjusting the resistance of the material electrically.