20230095092. DENOISING DIFFUSION GENERATIVE ADVERSARIAL NETWORKS simplified abstract (NVIDIA Corporation)

DENOISING DIFFUSION GENERATIVE ADVERSARIAL NETWORKS

Organization Name

NVIDIA Corporation

Inventor(s)

Zhisheng Xiao of Chicago IL (US)

Karsten Kreis of Vancouver (CA)

Arash Vahdat of Mountain View CA (US)

DENOISING DIFFUSION GENERATIVE ADVERSARIAL NETWORKS - A simplified explanation of the abstract

This abstract first appeared for US patent application 20230095092 titled 'DENOISING DIFFUSION GENERATIVE ADVERSARIAL NETWORKS

Simplified Explanation

The patent application describes a method and system for training and utilizing neural networks, specifically a denoising diffusion generative adversarial network (Denoising Diffusion GAN). This network reduces the number of denoising steps required during a reverse process, allowing for faster sample generation from noise.

• The Denoising Diffusion GAN does not assume a Gaussian distribution for large denoising steps.
• It applies a multi-model model to enable denoising with fewer steps.
• The system minimizes the divergence between the diffused real data distribution and the diffused generator distribution over multiple timesteps.

Potential applications of this technology:

• Faster sample generation from noise.
• Improved denoising processes.
• Enhanced training and utilization of neural networks.

Problems solved by this technology:

• Reduces the number of denoising steps required during a reverse process.
• Enables denoising with fewer steps.
• Minimizes divergence between real data and generator distributions.

Benefits of this technology:

• Faster and more efficient sample generation.
• Improved accuracy and effectiveness of denoising processes.
• Enhanced training and utilization of neural networks.

Original Abstract Submitted

apparatuses, systems, and techniques are presented to train and utilize one or more neural networks. a denoising diffusion generative adversarial network (denoising diffusion gan) reduces a number of denoising steps during a reverse process. the denoising diffusion gan does not assume a gaussian distribution for large steps of the denoising process and applies a multi-model model to permit denoising with fewer steps. systems and methods further minimize a divergence between a diffused real data distribution and a diffused generator distribution over several timesteps. accordingly, various embodiments may enable faster sample generation, in which the samples are generated from noise using the denoising diffusion gan.