Maximum-Likelihood Augmented Discrete Generative Adversarial Networks

By Tong Che, Yanran Li, Ruixiang Zhang, R. Devon Hjelm, Wenjie Li, Yangqiu Song, Yoshua Bengio. 2017

250 citations recorded by the source corpus.

Despite the successes in capturing continuous distributions, the application of generative adversarial networks (GANs) to discrete settings, like natural language tasks, is rather restricted. The fundamental reason is the difficulty of back-propagation through discrete random variables combined with the inherent instability of the GAN training objective. To address these problems, we propose Maximum-Likelihood Augmented Discrete Generative Adversarial Networks. Instead of directly optimizing the GAN objective, we derive a novel and low-variance objective using the discriminator's output that follows corresponds to the log-likelihood. Compared with the original, the new objective is proved to be consistent in theory and beneficial in practice. The experimental results on various discrete datasets demonstrate the effectiveness of the proposed approach.

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): “Che et al. (2017) instead optimize a lower-variance objective using the discriminator’s output, rather than the standard GAN objective.”
ColdGANs: Taming Language GANs with Cautious Sampling Strategies (2020, methodology): “Closer to our work, MaliGAN [7] proposes to rely on IS to consider an estimation of the data distribution as a target (via a KL objective).”
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].”
CausaLM: Causal Model Explanation Through Counterfactual Language Models (2020, background): “While in some cases, such as those presented in their paper, it might be plausible to generate counterfactual examples, in most cases in NLP it is still too hard to generate realistic texts with conditional generative models (Lin et al. 2017; Che et al. 2017; Rajeswar et al. 2017; Guo et al. 2018).”
Learning to Write with Cooperative Discriminators (2018, background): “However, backpropagating through discrete sequences and the inherent instability of the training objective (Che et al., 2017) both present significant challenges.”
Texygen: A Benchmarking Platform for Text Generation Models (2018, methodology): “Since then, many methods about improving Seq-GAN are proposed to further develop SeqGAN in many aspects; examples include gradient vanishing (MaliGAN [3], RankGAN [10], BRA in LeakGAN [5]), long-term robustness (LeakGAN).”
StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks (2017, background): “Recently, GANs have shown great potential in simulating complex data distributions, such as those of texts [6], images [32] and videos [48].”
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; Hjelm et al., 2017; Che et al., 2017).”
Synthesizing Tabular Data using Generative Adversarial Networks (2018, background): “However, generating discrete variables is a challenge for GANs: the authors in [26, 7] attempt a differential model by designing special functions or modifying the loss function; other researchers [50, 14, 46] use a reinforcement learning framework to train the non-differentiable model, making…”
Boundary-Seeking Generative Adversarial Networks (2017, background): “This distribution was discovered independently from us by Che et al. (2017) in the context of language modeling.”
Long Text Generation via Adversarial Training with Leaked Information (2017, methodology): “MaliGAN (Che et al. 2017) modifies the orginal GAN objective and proposes a set of training techniques to reduce the potential variance.”
Language GANs Falling Short (2018, background): “Notably, RankGAN (Lin et al., 2017), MaliGAN (Che et al., 2017), TextGAN (Zhang et al., 2017), LeakGAN (Guo et al., 2017) and IRL-GAN (Shi et al., 2018) were proposed soon after.”
Adversarial Generation of Natural Language (2017, methodology): “…2, the discrimi-nator is trained to discriminate between real and fake samples and the generator is trained to fool We investigate recurrent architectures as in (Lamb et al., 2016; Yu et al., 2016; Che et al., 2017) and convolutional architectures in both the generator as well as the discriminator.”
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).”
A Discriminator Improves Unconditional Text Generation without Updating the Generator (2020, unknown): “The performance of SeqGAN, MaliGAN and RankGAN are taken from (Lu et al., 2018b).”
Training language GANs from Scratch (2019, background): “Despite alleviating problems somewhat, all of the above methods require pre-training, sometimes together with teacher forcing [17] or interleaved supervised and adversarial training [15].”
How Generative Adversarial Networks and Its Variants Work : An Overview of GAN (2017, background): “MaliGAN [15] and BSGAN [44] take a similar approach in that both are inspired from pdata(x) = D∗(x) 1−D∗(x)pθ(x) for the optimal discriminator D ∗ and estimate the true distribution as p̃data(x) = 1 Z pθ(x) D(x) 1−D(x) where Z is a normalization factor.”