Boundary-Seeking Generative Adversarial Networks

By R. Devon Hjelm, Athul Paul Jacob, Tong Che, Kyunghyun Cho, Yoshua Bengio. 2017

176 citations recorded by the source corpus.

Generative adversarial networks (GANs) are a learning framework that rely on training a discriminator to estimate a measure of difference between a target and generated distributions. GANs, as normally formulated, rely on the generated samples being completely differentiable w.r.t. the generative parameters, and thus do not work for discrete data. We introduce a method for training GANs with discrete data that uses the estimated difference measure from the discriminator to compute importance weights for generated samples, thus providing a policy gradient for training the generator. The importance weights have a strong connection to the decision boundary of the discriminator, and we call our method boundary-seeking GANs (BGANs). We demonstrate the effectiveness of the proposed algorithm with discrete image and character-based natural language generation. In addition, the boundary-seeking objective extends to continuous data, which can be used to improve stability of training.

Subfields: Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

MaskGAN: Better Text Generation via Filling in the ______ (2018, methodology): “Hjelm et al. (2017) proposes an algorithmic solution and uses a boundary-seeking GAN objective along with importance sampling to generate text.”
Progressive Growing of GANs for Improved Quality, Stability, and Variation (2017, methodology): “Originally, Jensen-Shannon divergence was used as a distance metric (Goodfellow et al., 2014), and recently that formulation has been improved (Hjelm et al., 2017) and a number of more stable alternatives have been proposed, including least squares (Mao et al., 2016b), absolute deviation with…”
Improved Training of Wasserstein GANs (2017, methodology): “Other attempts at language modeling with GANs [31, 14, 29, 5, 15, 10] typically use discrete models and gradient estimators [27, 12, 16].”
Learning deep representations by mutual information estimation and maximization (2018, neutral): “Our results (Figure 15) show highly realistic images qualitatively competitive to other methods (Gulrajani et al., 2017; Hjelm et al., 2018).”
Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields (2017, methodology): “This data-set has previously been utilized to evaluate GANs [20].”
Which Training Methods for GANs do actually Converge? (2018, background): “(Hjelm et al., 2017) or use advanced normalization layers to stabilize the training (Miyato et al., 2018).”
GANs Trained by a Two Time-Scale Update Rule Converge to a Nash Equilibrium (2017, methodology): “The Large-scale CelebFaces Attributes (CelebA) dataset [37] has been used to evaluate GANs [24].”
Style Transfer from Non-Parallel Text by Cross-Alignment (2017, background): “Adversarial training over discrete samples Recently, a wide range of techniques addresses challenges associated with adversarial training over discrete samples generated by recurrent networks (Yu et al., 2016; Lamb et al., 2016; Che et al., 2017; Hjelm et al., 2017).”
Maximum-Likelihood Augmented Discrete Generative Adversarial Networks (2017, methodology): “The procedure was discovered independently from us by Hjelm et al. (2017) in the context of image generation.”
Robust Imitation of Diverse Behaviors (2017, background): “Recent work has focused on alleviating mode collapse in image modeling [2, 4, 19, 25, 41, 11], but so far these have not been demonstrated in the control domain.”
NetGAN: Generating Graphs via Random Walks (2018, background): “Other approaches modify the GAN objective to tackle the same challenge (Che et al., 2017; Hjelm et al., 2017).”
RankGAN: A Maximum Margin Ranking GAN for Generating Faces (2018, background): “The boundary-seeking GAN (BGAN) [11] extended GANs for discrete data while improving training stability for continuous data.”
Yes, we GAN: Applying Adversarial Techniques for Autonomous Driving (2019, methodology): “General functionality Vanilla GAN [2], Conditional GAN (CGAN) [9] Stabilizing functionality Wasserstein GAN (WGAN) [6], Improved WGAN (iWGAN) [7], Boundary-Seeking GAN (BGAN) [14]”
A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications (2020, methodology): “[452] introduced an algorithm for training GANs with discrete data that utilizes the estimated difference measure from the discriminator to compute importance weights for generated samples, and thus providing a policy gradient for training the generator.”
Adversarially Regularized Autoencoders (2017, methodology): “Recent work on generative adversarial networks (GANs) for text has mostly focused on dealing with the non-differentiable objective either through policy gradient methods (Che et al., 2017; Hjelm et al., 2018; Yu et al., 2017) or with the Gumbel-Softmax distribution (Kusner & Hernandez-Lobato, 2016).”
Learning Independent Features with Adversarial Nets for Non-linear ICA (2017, background): “Other recent work extended this interpretation of GANs to other di-vergences and distances between distributions (Arjovsky et al., 2017; Hjelm et al., 2017; Mao et al., 2016).”
Language Generation with Recurrent Generative Adversarial Networks without Pre-training (2017, methodology): “Gulrajani et al. (2017) and Hjelm et al. (2017) trained a purely generative adversarial model (without pre-training) for character-level sentence generation.”
How Generative Adversarial Networks and Their Variants Work (2017, methodology): “Maximum-likelihood augmented discrete GAN (MLADGAN) [14] and boundary-seeking GAN (BSGAN) [44] both take a policy gradient algorithm to generate discrete structured data.”
Generating and designing DNA with deep generative models (2017, background): “Only recently have GAN models been proposed which can successfully generate nontrivial natural language sequences [17–19, 16], though these”