TRaining
By embedding graphical encodings and using equal binning to label images, the training of cGANs is enhanced, addressing the challenge of sparse data and improving the generation of high-quality images.
Patent Information
- Authority / Receiving Office
- GB · GB
- Patent Type
- Applications
- Current Assignee / Owner
- ROLLS ROYCE PLC
- Filing Date
- 2025-06-20
- Publication Date
- 2026-04-22
AI Technical Summary
Conditional Generative Adversarial Networks (cGANs) face challenges in training accuracy when labelled training data is sparse, hindering their learning process in applications where data augmentation is necessary.
A method is provided for training a cGAN by embedding graphical encodings of physical parameter values into images, optimizing the generator and discriminator using a loss function that measures the difference in these encodings, and employing equal binning to label images based on physical output parameter values, thereby enhancing the training process.
The method improves the training efficiency and accuracy of cGANs by effectively utilizing sparse data, allowing for the generation of high-quality images that mimic the input data, as demonstrated by the comparison with thermo-mechanical simulation results.
Smart Images
Figure 00000000_0000_ABST
Abstract
Citation Information
Patent Citations
Design of engineering components
EP4177829A1
Design of engineering components
EP4177830A1