Reproduce “CINDES: Classification induced neural density estimator and simulator” 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
Dehao Dai, Jianqing Fan, Yihong Gu, Debarghya Mukherjee
Researchers from Princeton University developed CINDES, a deep neural network framework that rephrases density estimation as a classification problem. The method provides theoretical guarantees for adaptive, curse-of-dimensionality-free convergence rates when underlying densities exhibit low-dimensional structures, while also demonstrating superior empirical performance in both explicit and implicit density estimation across various benchmarks and real-world data.
Key insights
- Density estimation can be effectively converted into a standard classification problem, leveraging well-optimized deep learning tools for implementation ease.
- Deep neural networks can provably adapt to unknown low-dimensional structures (e.g., factorizable, compositional) within high-dimensional data, achieving faster convergence rates.
- The explicit CINDES density estimator can be naturally integrated into score-based diffusion models to achieve provably faster convergence for implicit sample generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.