Generative Adversarial Nets

By Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron C. Courville, Yoshua Bengio. Neural Information Processing Systems, 2014

2,369 citations recorded by the source corpus.

We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.

Subfields: Generative Models

View Paper: https://arxiv.org/abs/1406.2661

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

Citations in the corpus

Data-Efficient Image Recognition with Contrastive Predictive Coding (2019, background): “ntations that resemble simple cells in the brain [48]. The idea was later formalized in the context of probabilistic generative models [27], which have become a staple of modern unsupervised learning [28,22,35,56]. A key problem that generative models must overcome, however, is that not all low-leve”
Contrastive Multiview Coding (2019, methodology): “Representations better suited to such tasks have been learnt using deep neural networks, starting with seminal techniques such as Boltzmann machines [71, 65], autoencoders [30], variational autoencoders [40], generative adversarial networks [24] and autoregressive models [56].”
Dual Contradistinctive Generative Autoencoder (2020, neutral): “One benefit of having an AE/VAE framework compared with just a decoder as in GAN [16] is to be able to directly obtain the latent representation from the input images.”
Semantic Image Synthesis With Spatially-Adaptive Normalization (2019, background): “ontrollable, diverse outputs as shown in Figure1. 2. Related Work Deep generative models can learn to synthesize randomly sampled images. Recent methods include generative adversarial networks (GANs) [12] and variational autoencoder (VAE) [22]. Our work is built on GANs but aims for the conditional ”
Learning deep representations by mutual information estimation and maximization (2018, background): “heir respective joint distributions (which typically increases the marginal entropy in Equation 1) or minimizes the reconstruction error (Chen et al., 2016). In generative adversarial networks (GANs, Goodfellow et al., 2014), the representation of the discriminator has been shown to be suitable for ”
Stochastic Hamiltonian Gradient Methods for Smooth Games (2020, background): “, 2019), generative adversarial networks (Goodfellow et al., 2014), and some formulations in reinforcement learning (Pfau & Vinyals, 2016), empirical risk minimization yields finite sums of the form of (3).”
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (2016, methodology): “stinct desirable characteristics of the recovered SR image. The individual loss functions are described in more detail in Section2.2. 2.1. Adversarial network architecture Following Goodfellow et al. [22] we further deﬁne a discriminator network D D which we optimize in an alternating manner along ”
Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks (2015, methodology): “Generative Adversarial Networks (Goodfellow et al., 2014) generated images suffering from being noisy and incomprehensible.”
Unsupervised Domain Adaptation by Backpropagation (2014, methodology): “An idea related to ours is described in (Goodfellow et al., 2014).”
Rewriting a Deep Generative Model (2020, background): “Recently, thanks to rapid advances of deep generative models [26,45,31], learning-based image synthesis and editing methods have become widely-used tools in the community, enabling applications such as manipulating the semantics of an input scene [60,6,72,16], image colorization [83,34,52,85], photo”
Analysis Methods in Neural Language Processing: A Survey (2018, background): “(2018c) used generative adversarial networks (GANs) (Goodfellow et al., 2014) to minimize the distance between latent representations of input and adversarial examples, and performed perturbations in latent space.”
AutoAugment: Learning Augmentation Policies from Data (2018, methodology): “ablation experiments. AutoAugment vs. other automated data augmentation methods: Most notable amongst many previous data augmentation methods is the work of [47]. The setup in [47] is similar to GANs [18]: a generator learns to propose augmentation policy (a sequence of image processing operations) ”
Neural Approaches to Conversational AI (2018, neutral): “ “gaming” may occur unbeknownst to the system developer. If a generation system is optimized directly on a trainable metric, then the system and the metric become akin to an adversarial pair in GANs (Goodfellow et al., 2014), where the only goal of the generation system (Generator) is to fool the me”
Texygen: A Benchmarking Platform for Text Generation Models (2018, background): “The success of Generative Adversarial Network (GAN) [4] has inspired people to investigate adversarial training over textual, discrete data.”
Mitigating Unwanted Biases with Adversarial Learning (2018, methodology): “This is part of the network corresponds to the discriminator in a typical GAN [4].”
Video (language) modeling: a baseline for generative models of natural videos (2014, background): “ spaces can be difﬁcult, particularly the distribution of natural images which is highly concentrated and multimodal. Although some recent works have shown some progress on smaller resolution images (Goodfellow et al., 2014; Zoran & Weiss, 2012; Ranzato et al., 2013; Theis et al., 2011) using ge”
Inverting Visual Representations with Convolutional Networks (2015, methodology): “ To generate images we use the up-convolutional architecture, proposed by Dosovitskiy et al. [5]. Similar architectures are also often called “deconvolutiona” and have been used for generative models [19] and semantic segmentation [20]. 2 Network We base our study on the pre-trained version of the A”
Ensemble Adversarial Training: Attacks and Defenses (2017, methodology): “Adversarial training (Szegedy et al., 2013; Goodfellow et al., 2014b; Kurakin et al., 2017b; Madry et al., 2017) appears to hold the greatest promise for learning robust models.”
Recent advances in convolutional neural networks (2015, methodology): “q(x ) p(x )+ q(x ) 2 (27) By minimizing the JSD, we can make the two distributions p(x ) and q(x ) as close as possible. JSD has been successfully used in the Generative Adversarial Networks (GANs) [87, 96, 97]. In contrast to VAEs that model the relationship between x and z directly, GANs are expl”