Research map / Architectures and efficiency / RNNs and normalization
RNNs Incrementally Evolving on an Equilibrium Manifold: A Panacea for Vanishing and Exploding Gradients?
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 |
|---|---|---|---|
| Replace the vanilla recurrent update with ERNN equilibrium-manifold updates | RNNs · HAR-2 | accuracy (%): 91.31 -> 96.30 | Helped |
| Replace the vanilla recurrent update with ERNN equilibrium-manifold updates | RNNs · DSA-19 | accuracy (%): 71.68 -> 88.11 | Helped |
| Use ERNN instead of FastGRNN-LSQ | recurrent models · Pixel-MNIST | accuracy (%): 98.72 -> 98.13 | Hurt |
| Use ERNN instead of FastGRNN-LSQ | recurrent models · Yelp-5 | accuracy (%): 59.51 -> 59.34 | Hurt |
| Increase fixed-point iterations from K=1 to K=5 (ablation) | ERNN · HAR-2 | accuracy (%): 95.32 -> 96.30 | Helped |
| Increase fixed-point iterations from K=1 to K=5 (ablation) | ERNN · DSA-19 | accuracy (%): 88.11 -> 87.37 | Hurt |
| Increase fixed-point iterations from K=1 to K=5 (ablation) | ERNN · Google-12 | accuracy (%): 93.93 -> 94.71 | Helped |
| Increase fixed-point iterations from K=1 to K=5 (ablation) | ERNN · Google-30 | accuracy (%): 93.77 -> 94.23 (K=3: 91.30) | Mixed |
| Increase fixed-point iterations from K=1 to K=3 (ablation) | ERNN · Pixel-MNIST | accuracy (%): 97.73 -> 98.13 | Helped |
| Increase fixed-point iterations from K=1 to K=3 (ablation) | ERNN · Yelp-5 | accuracy (%): 58.16 -> 59.34 | Helped |
| Replace ReLU with Sigmoid (ablation) | ERNN (HAR-2, K=1) · HAR-2 | accuracy (%): 95.32 -> 92.16 | Hurt |
| Replace ReLU with Sigmoid (ablation) | ERNN (HAR-2, K=3) · HAR-2 | accuracy (%): 95.52 -> 93.35 | Hurt |
Closest papers by meaning
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.90)
- Skip RNN: Learning to Skip State Updates in Recurrent Neural Networks (ICLR 2018 · similarity 0.90)
- UnICORNN: A recurrent model for learning very long time dependencies (ICML 2021 · similarity 0.90)
- Time Adaptive Recurrent Neural Network (CVPR 2021 · similarity 0.89)
- AntisymmetricRNN: A Dynamical System View on Recurrent Neural Networks (ICLR 2019 · similarity 0.89)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · 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.