Research map / Probabilistic and scientific modeling / Variational and generative sampling
The continuous categorical: a novel simplex-valued exponential family
Elliott Gordon-Rodriguez, Gabriel Loaiza-Ganem, John Cunningham
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the CC likelihood instead of the Dirichlet likelihood | Linear regression model · UK election data test set | MAE: 0.105 -> 0.075 | Improved |
| Use the CC likelihood instead of the Dirichlet likelihood | Linear regression model · UK election data test set | RMSE: 0.132 -> 0.101 | Improved |
| Use the CC likelihood instead of the Dirichlet likelihood | MLP regression model · UK election data test set | MAE: 0.087 -> 0.072 | Improved |
| Use the CC likelihood instead of the Dirichlet likelihood | MLP regression model · UK election data test set | RMSE: 0.112 -> 0.097 | Improved |
| Use the CC objective instead of Soft XE | Student network · MNIST test set | accuracy: 94.9% -> 95.6% | Improved |
| Use the CC objective instead of Soft XE | Student network · MNIST test set | RMSE: 0.029 -> 0.024 | Improved |
| Use teacher soft targets with Soft XE instead of original hard labels | Student network · MNIST test set | accuracy: 93.2% -> 94.9% | Improved |
| Increase the temperature of teacher soft targets (ablation) | Dirichlet student model · MNIST | test error | Improved |
| Add L2 regularization to the regression network (ablation) | Dirichlet UK election regression model · UK election data | test error | Mixed |
| Replace vanilla LDA with a CC-prior topic model | Topic model · NeurIPS papers held-out | perplexity per word: 557 -> 562 | Hurt |
| Fit a CC mean estimator instead of a Dirichlet mean estimator | Simplex-valued data estimator · Simulation study | empirical bias | Improved |
| Use the CC objective instead of the Dirichlet likelihood | Student network · MNIST test set | accuracy: 90.6% -> 95.6% | Improved |
Closest papers by meaning
- A New Distribution on the Simplex with Auto-Encoding Applications (NeurIPS 2019 · similarity 0.87)
- Categorical Normalizing Flows via Continuous Transformations (ICLR 2021 · similarity 0.86)
- Sparse Continuous Distributions and Fenchel-Young Losses (JMLR 2022 · similarity 0.86)
- Unified Discrete Diffusion for Categorical Data (JMLR 2025 · similarity 0.86)
- Spherical Flows for Sampling Categorical Data (NeurIPS 2026 · similarity 0.84)
- Multimarginal Generative Modeling with Stochastic Interpolants (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
