InvNet: Encoding Geometric and Statistical Invariances in Deep Generative Models

• DOI: 10.1609/aaai.v34i04.5863
• Publication Date: 2020-06-29T21:30:20Z
• Authors: Ameya Joshi; Minsu Cho; Viraj Shah; Balaji Pokuri; Soumik Sarkar; Baskar Ganapathys...; Chinmay Hegde
• Keywords: Generative model

Generative Adversarial Networks (GANs), while widely successful in modeling complex data distributions, have not yet been sufficiently leveraged scientific computing and design. Reasons for this include the lack of flexibility GANs to represent discrete-valued image data, as well control over physical properties generated samples. We propose a new conditional generative approach (InvNet) that efficiently enables images, allowing their parameterized geometric statistical properties. evaluate our on several synthetic real world problems: navigating manifolds shapes with desired sizes; generation binary two-phase materials; (challenging) problem generating multi-orientation polycrystalline microstructures.