Research map / Probabilistic and scientific modeling / Variational and generative sampling
Z-Forcing: Training Stochastic Recurrent Networks
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 |
|---|---|---|---|
| Add the auxiliary cost to Ours (ablation) | Z-Forcing · Blizzard test | average log-likelihood per sequence: ≥ 14435 -> ≥ 15430 | Helped |
| Add the auxiliary cost to Ours (ablation) | Z-Forcing · TIMIT test | average log-likelihood per sequence: ≥ 68132 -> ≥ 69530 | Helped |
| Add KL annealing to Ours (ablation) | Z-Forcing · Blizzard test | average log-likelihood per sequence: ≥ 14435 -> ≥ 14226 | Hurt |
| Add KL annealing to Ours (ablation) | Z-Forcing · TIMIT test | average log-likelihood per sequence: ≥ 68132 -> ≥ 68903 | Helped |
| Add KL annealing alongside the auxiliary cost (ablation) | Z-Forcing · TIMIT test | average log-likelihood per sequence: ≥ 69530 -> ≥ 70469 | Helped |
| Add the auxiliary cost (ablation) | Z-Forcing · Sequential MNIST test | negative log-likelihood: ≤ 80.60 -> ≤ 80.09 | Helped |
| Use KL annealing (ablation) | Z-Forcing · Sequential MNIST | negative log-likelihood | Hurt |
| Add the auxiliary cost with α=0.0025 (ablation) | Z-Forcing · IMDB valid | word perplexity (ELBO): 53.93 -> 55.71 | Hurt |
| Add the auxiliary cost with α=0.0025 (ablation) | Z-Forcing · IMDB test | word perplexity (ELBO): 54.67 -> 56.57 | Hurt |
| Increase auxiliary-cost hyperparameter α from 0.0025 to 0.005 (ablation) | Z-Forcing with auxiliary cost · IMDB valid | word perplexity (ELBO): 55.71 -> 65.03 | Hurt |
| Increase auxiliary-cost hyperparameter α from 0.0025 to 0.005 (ablation) | Z-Forcing with auxiliary cost · IMDB test | word perplexity (ELBO): 56.57 -> 65.84 | Hurt |
| Train with the IWAE bound instead of the standard ELBO (ablation) | Z-Forcing · IMDB validation | perplexity | No clear effect |
Closest papers by meaning
- Sequential Neural Models with Stochastic Layers (NeurIPS 2016 · similarity 0.90)
- Re-examination of the Role of Latent Variables in Sequence Modeling (NeurIPS 2019 · similarity 0.90)
- Fast-Slow Recurrent Neural Networks (NeurIPS 2017 · similarity 0.88)
- Inferring stochastic low-rank recurrent neural networks from neural data (NeurIPS 2024 · similarity 0.88)
- Variable Computation in Recurrent Neural Networks (ICLR 2017 · similarity 0.88)
- Stable Recurrent Models (ICLR 2019 · similarity 0.88)
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.