Reproduce “Nonparametric Generative Modeling with Conditional Sliced-Wasserstein Flows” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chao Du, Tianbo Li, Tianyu Pang, Shuicheng Yan, Min Lin
A nonparametric generative modeling approach called Conditional Sliced-Wasserstein Flows (CSWF) is introduced by Sea AI Lab, enabling high-quality conditional and unconditional image synthesis. The method incorporates visual inductive biases and achieves performance levels competitive with many deep parametric models, without relying on backpropagation.
Key insights
- Simulating joint Sliced-Wasserstein Flows on a combined (data, condition) space can effectively approximate conditional generative modeling through velocity masking.
- Incorporating domain-specific inductive biases, such as locally-connected projections and pyramidal schedules, is crucial for nonparametric methods to achieve high-quality results on structured data like images.
- Nonparametric generative models, when appropriately enhanced with conditional mechanisms and domain-specific biases, can achieve performance competitive with deep parametric models without relying on backpropagation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.