Research map / Probabilistic and scientific modeling / Variational and generative sampling
Efficient Marginalization of Discrete and Structured Latent Variables via Sparsity
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace dense softmax marginalization with sparsemax marginalization | Emergent communication model · Emergent communication test | communication success (%): 93.37 ± 0.42 -> 93.35 ± 0.50 | No clear effect |
| Replace dense softmax marginalization with sparsemax marginalization | Emergent communication model · Emergent communication test | decoder calls: 256 -> 3.13 ± 0.48 | Helped |
| Use exact marginalization instead of stochastic gradient estimation | Emergent communication model · Emergent communication test | communication success (%): 55.36 ± 2.92 -> 93.35 ± 0.50 | Helped |
| Use top-k sparsemax rather than a stochastic estimator | Bit-vector VAE · Fashion-MNIST, D=32 | rate-distortion performance | No clear effect |
| Use sparse marginalization rather than an unbiased stochastic estimator | Bit-vector VAE · Fashion-MNIST, D=128 | rate-distortion performance | Helped |
| Use sparse marginalization rather than SFE | Bit-vector VAE · Fashion-MNIST | estimated negative log-likelihood | Helped |
| Use Gumbel-Softmax instead of sparse marginalization | Semisupervised VAE · MNIST | ELBO and accuracy | Hurt |
Closest papers by meaning
- Generative Marginalization Models (ICML 2024 · similarity 0.89)
- Beyond Softmax: A Natural Parameterization for Categorical Random Variables (ICML 2026 · similarity 0.89)
- Leveraging Recursive Gumbel-Max Trick for Approximate Inference in Combinatorial Spaces (NeurIPS 2021 · similarity 0.88)
- Marginalization is not Marginal: No Bad VAE Local Minima when Learning Optimal Sparse Representations (ICML 2023 · similarity 0.88)
- Learning latent variable structured prediction models with Gaussian perturbations (NeurIPS 2018 · similarity 0.87)
- SUMO: Unbiased Estimation of Log Marginal Probability for Latent Variable Models (ICLR 2020 · similarity 0.87)
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.