• Yoga Harshitha Duddukuri

This paper discovers the usage and studies the effect of diffusion based generation model to transform the raw training data of GAN into standard normal latent space. This improved the training stability and reduced resource utilization and played a crucial role in decreasing the training error. Papers till date have explored the use of generating models for generating the data it is trained on. The research done on natural language processing shows the significance of word embedding and transforming unstructured data into trainable domain [1] , which involve in lot of resources for training, and there are very few papers on generalizing the process, as compared to the papers that exists on its applications. In this paper we are using diffusion based generation model as intermediate processing block on USDA food database. This offers insights into extending the application of GAN networks into fields that did not gain much attention till date. The main application is to prepare a meal plan by generating various combinations of dishes in the database so as to meet the nutritional needs. GAN that is trained directly on the food database was unstable and generated a huge error. The standard normalizing and de normalizing techniques were barely up to the help in decreasing the error. Instead of training the GAN to directly generate the data, we have stacked already trained diffusion based generation model at the end of GAN, that was trained to map the food data into latent space, and trained GAN network for time series generation of different food combinations that would meet the nutritional requirements.

diffusion based generation, improving the training of GAN, GAN applications, normalization, embedding vectors, data processing.

"Deploying Diffusion Based Generation Model For Improved Training Of GAN", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.10, Issue 9, page no.e522-e527, September-2023, Available :http://www.jetir.org/papers/JETIR2309463.pdf

"Deploying Diffusion Based Generation Model For Improved Training Of GAN", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.10, Issue 9, page no. ppe522-e527, September-2023, Available at : http://www.jetir.org/papers/JETIR2309463.pdf