Research map / Architectures and efficiency / RNNs and normalization
Adding Attentiveness to the Neurons in Recurrent Neural Networks
Pengfei Zhang, Jianru Xue, Cuiling Lan, Wenjun Zeng, Zhanning Gao, Nanning Zheng
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 |
|---|---|---|---|
| add EleAttG to standard RNN | 1-layer sRNN · NTU skeleton CS | accuracy: 51.6 -> 61.6 | Improved |
| add EleAttG to LSTM | 3-layer LSTM · NTU skeleton CS | accuracy: 77.2 -> 78.4 | Improved |
| add EleAttG to LSTM | 3-layer LSTM · NTU skeleton CV | accuracy: 83.0 -> 85.0 | Improved |
| add EleAttG to GRU | 3-layer GRU · NTU skeleton CS | accuracy: 75.2 -> 79.8 | Improved |
| add EleAttG to GRU | 3-layer GRU · NTU skeleton CV | accuracy: 81.5 -> 87.1 | Improved |
| add EleAttG to GRU on RGB CNN features | GRU · JHMDB RGB average | accuracy: 60.9 -> 62.9 | Improved |
| add EleAttG to GRU on RGB CNN features | GRU · NTU RGB CV | accuracy: 66.8 -> 70.6 | Improved |
| add skeleton rotation augmentation (ablation) | EleAtt-GRU · NTU skeleton CV | accuracy: 87.1 -> 88.4 | Improved |
| add EleAttG to GRU | 3-layer GRU · N-UCLA | accuracy: 84.3 -> 89.0 | Improved |
| add EleAttG to GRU | 3-layer GRU · SYSU CS | accuracy: 82.1 -> 84.9 | Improved |
| use sigmoid (no sum-to-1) instead of softmax in EleAttG (ablation) | EleAttG in 2nd GRU layer · NTU skeleton CS | accuracy: 72.7 -> 77.3 | Improved |
| add EleAttG with softmax constraint to 2nd layer (ablation) | GRU baseline · NTU skeleton CS | accuracy: 75.2 -> 72.7 | Hurt |
Closest papers by meaning
- Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks (ICML 2018 · similarity 0.86)
- SkipW: Resource Adaptable RNN with Strict Upper Computational Limit (ICLR 2021 · similarity 0.86)
- Gating Revisited: Deep Multi-layer RNNs That Can Be Trained (TPAMI 2021 · similarity 0.85)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.84)
- Structured flexibility in recurrent neural networks via neuromodulation (NeurIPS 2024 · similarity 0.84)
- Exploring the Promise and Limits of Real-Time Recurrent Learning (ICLR 2024 · similarity 0.84)
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.
