Research map / Architectures and efficiency / RNNs and normalization
Noisin: Unbiased Regularization for Recurrent Neural 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 multiplicative Bernoulli noise to an LSTM | LSTM · Penn Treebank medium test | perplexity: 77.0 -> 71.4 | Helped |
| Add multiplicative Bernoulli noise to an LSTM | LSTM · Penn Treebank large test | perplexity: 75.3 -> 68.3 | Helped |
| Add multiplicative Beta noise to a dropout-LSTM | Dropout-LSTM · Penn Treebank medium test | perplexity: 77.0 -> 69.2 | Helped |
| Add multiplicative Gaussian noise to a dropout-LSTM | Dropout-LSTM · Penn Treebank large test | perplexity: 75.3 -> 66.1 | Helped |
| Add multiplicative Bernoulli noise to an LSTM | LSTM · Wikitext-2 medium test | perplexity: 84.8 -> 86.6 | Helped |
| Add multiplicative Laplace noise to a dropout-LSTM | Dropout-LSTM · Wikitext-2 medium test | perplexity: 84.8 -> 82.1 | Helped |
| Add multiplicative Beta noise to an LSTM | LSTM · Wikitext-2 large test | perplexity: 91.0 -> 82.9 | Helped |
| Add multiplicative Bernoulli noise to a dropout-LSTM | Dropout-LSTM · Wikitext-2 large test | perplexity: 91.0 -> 76.8 | Helped |
| Add Noisin to AWD-LSTM-MoS with dynamic evaluation | AWD-LSTM-MoS + dynamic evaluation · Penn Treebank test | perplexity: 47.7 -> 47.6 | No clear effect |
| Use additive rather than multiplicative noise (ablation) | LSTM with Noisin · Penn Treebank and Wikitext-2 | perplexity | Hurt |
| Vary the noise distribution while controlling its variance (ablation) | LSTM with Noisin · Penn Treebank and Wikitext-2 | perplexity | Mixed |
Closest papers by meaning
- Data Noising as Smoothing in Neural Network Language Models (ICLR 2017 · similarity 0.89)
- Skip RNN: Learning to Skip State Updates in Recurrent Neural Networks (ICLR 2018 · similarity 0.89)
- Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activations (ICLR 2017 · similarity 0.89)
- Variational Smoothing in Recurrent Neural Network Language Models (ICLR 2019 · similarity 0.88)
- State-Regularized Recurrent Neural Networks (ICML 2019 · similarity 0.88)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · similarity 0.86)
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.