TreeGAN: Syntax-Aware Sequence Generation with Generative Adversarial Networks

Abstract

Generative Adversarial Networks (GANs) have shown great capacity on image generation, in which a discriminative model guides the training of a generative model to construct images that resemble real images. Recently, GANs have been extended from generating images to generating sequences (e.g., poems, music and codes). Existing GANs on sequence generation mainly focus on general sequences, which are grammar-free. In many real-world applications, however, we need to generate sequences in a formal language with the constraint of its corresponding grammar. For example, to test the performance of a database, one may want to generate a collection of SQL queries, which are not only similar to the queries of real users, but also follow the SQL syntax of the target database. Generating such sequences is highly challenging because both the generator and discriminator of GANs need to consider the structure of the sequences and the given grammar in the formal language. To address these issues, we study the problem of syntax-aware sequence generation with GANs, in which a collection of real sequences and a set of pre-defined grammatical rules are given to both discriminator and generator. We propose a novel GAN framework, namely TreeGAN, to incorporate a given Context-Free Grammar (CFG) into the sequence generation process. In TreeGAN, the generator employs a recurrent neural network (RNN) to construct a parse tree. Each generated parse tree can then be translated to a valid sequence of the given grammar. The discriminator uses a tree-structured RNN to distinguish the generated trees from real trees. We show that TreeGAN can generate sequences for any CFG and its generation fully conforms with the given syntax. Experiments on synthetic and real data sets demonstrated that TreeGAN significantly improves the quality of the sequence generation in context-free languages.

Spectrum of Advancements and Developments in Multidisciplinary Domains for Generative Adversarial Networks (GANs)

“…information diffusion technology (idt), the heuristic mega-trend-diffusion (hmtd) technology and the bootstrap technique Advantage: scalable to perform forecast for demand of electricity and the traffic supply Luo et al [52] 2018 A technique for predicting the prices of the crude oil using adap- [55] 2019 Double P-buried layers MISFET (DP-MISFET) is proposed Simulated and characteristics are analysed by the Sentaurus TCAD tool Road network generation and path planning Albert et al [56] 2018 Novel technique to simulate real like urban designs fine-tuned with urban land-use inventory Advantage: synthetic urban pattern is formulated to qualitatively regenerate the spatial structures perceived in urban designs Mohammadi et al [57] 2018 Precise and reliable paths for navigation software including wayfinding for disabled people, route identification for evacuations, robotic navigations for autonomous vehicles Advantage: high accuracy of the classification task with high quality of the generated paths is achieved Testing and validation Zhang et al [58] 2018 Unsupervised model for automatic verification and validation of the consistent behavior of autonomous vehicle driving systems Real time validation is also achieved Segura et al [59] 2016 Metamorphic verification and validation approach for identifying unusual behaviors of autonomous vehicle systems along with input validation Zhihui Li et al [60] 2019 Create fuzzing data using Wasserstein GANs (wgans) Advantage: does not require specification of input data Significant for testing of industrial control systems (icss) Software designing and development Li et al [61] 2019 Layoutgan-Wireframe designing i.e. layouts generation of relational graphic elements to wireframe images by modelling geometric relations of different types of two dimensional elements Advantage: introduction of wireframe rendering layer which produce a set of relational graphic controls Liu et al [62] 2018 Treegan for source code generation Advantage: syntax-aware sequence generation Fault prediction Gao et al [63] 2019 ASM1D-GAN a model to identify the faults by extracting features related to faults from real fault samples and create the similar one Advantage: integration of data creation and fault determination Zhou et al [64] 2019 Synthesize vibrational fault samples using a technique of global optimization Advantage: feature extraction of feature using limited number of samples and its effective representation using auto-encoder Filter the non-compliant synthetic samples which are not useful for reliable fault diagnosis Zheng et al [65] 2019 Gan-fp utilizes multiple GANs to create training samples and an inference network in parallel to predict failures for newly crafted samples Improved performance as well as significant socio-economic impact Text generation Subramanian et al [66] 2018 Ability to create sentence outlines using an adversarial model which learns the distribution of sentences in a hidden space persuaded by sentence encoder Advantage: produce real like samples with multinomial sampling Liang et al …”

Paraphrase Generation via Adversarial Penalizations

“…Instead of calculating the reward using the likelihood, most works rely on a discriminator (Yu et al, 2017;Li et al, 2018a;Liu et al, 2018;Guo et al, 2018) or an evaluator (Li et al, 2018b).…”

“…However, highlight the exposure bias problem (Bengio et al, 2015;Ranzato et al, 2015) in paraphrase generation. Some works address the problem by applying REINFORCE (Williams, 1992) to generate text in their adversarial setups (Yu et al, 2017;Fedus et al, 2018;Li et al, 2018a;Liu et al, 2018;de Masson d'Autume et al, 2019). However, simi-lar to prior works (He et al, 2019;, we observe that REINFORCE has a high variance and is difficult to tune.…”

GANCoder: An Automatic Natural Language-to-Programming Language Translation Approach Based on GAN

“…The discriminator is responsible for judging whether the ASTs generated by the generator are consistent with the natural language utterance semantics. We use GAN to improve the generative model G θ , the optimization equation of the GAN network is as follows [13]:…”

“…In the generator, the structure of the AST is generated recursively top-down and left-right, but in the discriminator, the entire AST is encoded bottom-up from leaf node to root nodes of AST, and the final vector is used as the semantic vector of the program fragment. In this way, the syntax and logic of the program fragment can be learned in a bottom-up manner [13,14]. Let h r be the final encoding vector for the entire abstract syntax tree.…”