Research map / Probabilistic and scientific modeling / Dynamical systems
Noisy Recurrent Neural Networks
Soon Hoe Lim, N. Benjamin Erichson, Liam Hodgkinson, Michael W. Mahoney
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 |
|---|---|---|---|
| noise injection (NRNN 0.02/0.05) vs deterministic Lipschitz RNN | Lipschitz RNN · Ordered pixel-by-pixel MNIST, white noise σ=0.3 | test accuracy (%): 47.1 -> 73.5 | Improved |
| noise injection (NRNN 0.02/0.05) vs Lipschitz RNN | Lipschitz RNN · Ordered pixel-by-pixel MNIST, S&P α=0.1 | test accuracy (%): 73.5 -> 85.5 | Improved |
| noise injection (NRNN 0.02/0.05) vs Lipschitz RNN | Lipschitz RNN · Ordered pixel-by-pixel MNIST, adversarial r=0.15 | test accuracy (%): 37.1 -> 70.6 | Improved |
| noise injection (NRNN 0.02/0.05) vs Lipschitz RNN | Lipschitz RNN · Ordered pixel-by-pixel MNIST, clean | test accuracy (%): 99.2 -> 99.1 | Hurt |
| increase additive noise from 0.02 to 0.05 (multiplicative 0.02) | NRNN · Ordered pixel-by-pixel MNIST, white noise σ=0.3 | test accuracy (%): 62.9 -> 73.5 | Improved |
| increase additive noise from 0.02 to 0.05 (multiplicative 0.02) | NRNN · Ordered pixel-by-pixel MNIST, adversarial r=0.15 | test accuracy (%): 58.3 -> 70.6 | Improved |
| increase additive noise from 0.02 to 0.05 (multiplicative 0.02) | NRNN · Ordered pixel-by-pixel MNIST, clean | test accuracy (%): 99.1 -> 99.1 | No clear effect |
| NRNN 0.02/0.02 vs Lipschitz RNN | Lipschitz RNN · Ordered pixel-by-pixel MNIST, white noise σ=0.2 | test accuracy (%): 78.9 -> 88.4 | Improved |
| NRNN 0.02/0.02 vs CoRNN | CoRNN · Ordered pixel-by-pixel MNIST, adversarial r=0.15 | test accuracy (%): 35.1 -> 58.3 | Improved |
| noise injection (NRNN 0.03/0.06) vs Lipschitz RNN | Lipschitz RNN · ECG, white noise σ=1.2 | test accuracy (%): 88.9 -> 92.6 | Improved |
| noise injection (NRNN 0.03/0.06) vs Lipschitz RNN | Lipschitz RNN · ECG, multiplicative noise σM=1.2 | test accuracy (%): 95.6 -> 96.5 | Improved |
| noise injection (NRNN 0.03/0.06) vs Lipschitz RNN | Lipschitz RNN · ECG, clean | test accuracy (%): 97.7 -> 97.7 | No clear effect |
Closest papers by meaning
- Stable Recurrent Models (ICLR 2019 · similarity 0.89)
- Noise Injection Node Regularization for Robust Learning (ICLR 2023 · similarity 0.89)
- Understanding Generalization in Recurrent Neural Networks (ICLR 2020 · similarity 0.88)
- Understanding and Controlling Memory in Recurrent Neural Networks (ICML 2019 · similarity 0.88)
- SBO-RNN: Reformulating Recurrent Neural Networks via Stochastic Bilevel Optimization (NeurIPS 2021 · similarity 0.88)
- Benign Oscillation of Stochastic Gradient Descent with Large Learning Rate (ICLR 2024 · 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.
