Research map / Probabilistic and scientific modeling / Variational and generative sampling
Adaptive Perturbation-Based Gradient Estimation for Discrete Latent Variable Models
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 |
|---|---|---|---|
| Adapt the finite-difference step size during training | IMLE · Synthetic categorical gradient estimation | Cosine similarity to the true gradient | Helped |
| Use central rather than forward differences (ablation) | AIMLE · BeerAdvocate aroma, K=5 | Subset precision: 44.668 -> 62.056 | Helped |
| Use central rather than forward differences (ablation) | AIMLE · Latent graph recovery, T=10 | ELBO: -1877.90 -> -2018.39 | Hurt |
| Use central rather than forward differences (ablation) | AIMLE · Latent graph recovery, T=20 | Edge precision: 40.48 -> 70.89 | Helped |
| Increase the number of AIMLE samples (ablation) | AIMLE · Discrete VAE on MNIST, K=1 | Test loss | Helped |
| Use AIMLE instead of Gumbel-Softmax | Discrete VAE on MNIST, K=1 · Discrete VAE, K=1 | Test loss | Hurt |
| Increase the AIMLE update step from 10^{-3} to 10^{-2} (ablation) | AIMLE · Warcraft shortest path prediction | Convergence rate: 10^{-3} -> 10^{-2} | Helped |
| Replace forward differences with central differences (ablation) | AIMLE · BeerAdvocate, all aspects | Subset precision | Helped |
Closest papers by meaning
- Implicit MLE: Backpropagating Through Discrete Exponential Family Distributions (NeurIPS 2021 · similarity 0.87)
- Gradient Estimation with Discrete Stein Operators (NeurIPS 2022 · similarity 0.86)
- Gradient Estimation for Binary Latent Variables via Gradient Variance Clipping (AAAI 2023 · similarity 0.86)
- Variational Inference for Sequential Data with Future Likelihood Estimates (ICML 2020 · similarity 0.85)
- Non-asymptotic Analysis of Biased Adaptive Stochastic Approximation (NeurIPS 2024 · similarity 0.85)
- Revisiting Zeroth-Order Optimization: Minimum-Variance Two-Point Estimators and Directionally Aligned Perturbations (ICLR 2025 · similarity 0.83)
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.