h-zhang00
commented
8 months ago After reading your paper, I am very curious about how the Inverse network achieves the one-to-many property-structure mapping. I believe that once the network is trained, the parameters are fixed, and given the same input, the output will always be the same. Can the Inverse network design different structures by providing the same set of SACs parameters? Thank you very much. I would appreciate it if you could address my issue.
Hi. One-to-many issue is not really our concern in this paper. Please refer to the following paper that employs VAE to perform inverse design in a probabilistically generative manner, which is a common approach to tackle the one-to-many problem.
"Probabilistic Representation and Inverse Design of Metamaterials Based on a Deep Generative Model with Semi-Supervised Learning Strategy"
https://onlinelibrary.wiley.com/doi/abs/10.1002/adma.201901111
After reading your paper, I am very curious about how the Inverse network achieves the one-to-many property-structure mapping. I believe that once the network is trained, the parameters are fixed, and given the same input, the output will always be the same. Can the Inverse network design different structures by providing the same set of SACs parameters? Thank you very much. I would appreciate it if you could address my issue.