Research map / Probabilistic and scientific modeling / Variational and generative sampling
Sparse Communication via Mixed Distributions
António Farinhas, Wilker Aziz, Vlad Niculae, Andre Martins
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 Gaussian-Sparsemax mixed messages instead of continuous Gumbel-Softmax messages | Emergent communication · Emergent communication test | success (%): 78.84 ± 8.07 -> 80.88 ± 0.50 | Mixed |
| Use Gaussian-Sparsemax mixed messages instead of Gumbel-Softmax messages | Emergent communication · Emergent communication test | nonzeros: 256 -> 1.57 ± 0.02 | Improved |
| Use Gaussian-Sparsemax mixed messages instead of Gumbel-Softmax ST messages | Emergent communication · Emergent communication test | success (%): 49.96 ± 9.51 -> 80.88 ± 0.50 | Improved |
| Use Gaussian-Sparsemax with exact entropy rather than estimated entropy (ablation) | Fashion-MNIST bit-vector VAE · Fashion-MNIST test | NLL (bits/dim): 3.53 -> 3.49 | Improved |
| Use Gaussian-Sparsemax instead of Hard Concrete | Fashion-MNIST bit-vector VAE · Fashion-MNIST test | NLL (bits/dim): 3.57 -> 3.49 | Improved |
| Use Mixed Dirichlet instead of Dirichlet | Mixed-latent VAE · MNIST test | NLL: 93.81 -> 106.59 | Hurt |
| Constrain the prior to the simplex rather than allow the full R^10 | Mixed-latent VAE · MNIST test | NLL: 91.12 -> 106.59 | Hurt |
Closest papers by meaning
- Discrete-Valued Neural Communication (NeurIPS 2021 · similarity 0.85)
- The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables (ICLR 2017 · similarity 0.85)
- Connecting Jensen–Shannon and Kullback–Leibler Divergences: A New Bound for Representation Learning (NeurIPS 2025 · similarity 0.85)
- Flow-based Variational Mutual Information: Fast and Flexible Approximations (ICLR 2025 · similarity 0.85)
- Synergy Between Sufficient Changes and Sparse Mixing Procedure for Disentangled Representation Learning (ICLR 2025 · similarity 0.85)
- Partial Information Decomposition via Normalizing Flows in Latent Gaussian Distributions (NeurIPS 2025 · 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.
