bharathreddy1997 Goto Github PK

block-chain-hashing

blockchain-bank-application

end-of-course-project-bachelors-

hubspot-challenge

Hubspot full-time coding challenge

jenkinssample

media-prediction

multiple-sleeping-barber-problem

pseudocode-converter-python-textx

rnn.wgan

Code for training and evaluation of the model from "Language Generation with Recurrent Generative Adversarial Networks without Pre-training"

sentigan-curriculum

sentiment-analysis-covid19

about Covid19

terrorist-data-visualization

testim

text-generation-using-gan-and-rnn

Automatically generating syntactically correct and semantically meaningful text has many applications in dialogue systems, image captioning, machine translation and more. Indeed, while the most advanced of today's rule-based chatbots may not be able to generate good conversation, AI systems based on machine learning can perform better. This paper describes and compares two methods for generating artificial text namely using GANs (Generative Adversarial Networks) and using RNNs (Recurrent Neural Networks). We apply these to three different domains namely communications between ATC (Air traffic control) and pilots, children's nursery rhymes and the poetry of the Irish poet Seamus Heaney. In each case we work with small vocabularies and small amounts of training data as that is our use case. Sample texts based on training data from each domain from the RNN and the GAN approaches are evaluated using BERT's score and in the case of Heaney's poetry we also include extensive feedback from survey respondents. Our findings are that realistic text can be generated in each domain even with small amounts of training data and we conclude that in such scenarios, RNN-generated text is better than GAN-generated, as well as being less computationally expensive.

weather-prediction-website