Research map / Architectures and efficiency / RNNs and normalization
Non-Local Recurrent Neural Memory for Supervised Sequence Modeling
Canmiao Fu, Wenjie Pei, Qiong Cao, Chaopeng Zhang, Yong Zhao, Xiaoyong Shen, Yu-Wing Tai
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 NRNM memory to the LSTM backbone | LSTM · NTU RGB+D (Cross-View) | accuracy %: 84.0 -> 89.2 | Improved |
| add NRNM memory to the LSTM backbone | LSTM · NTU RGB+D (Cross-Subject) | accuracy %: 70.3 -> 80.8 | Improved |
| add NRNM memory to the LSTM backbone | LSTM · IMDB | accuracy %: 89.8 -> 93.1 | Improved |
| increase NRNM block size up to 8 (ablation) | NRNM · NTU RGB+D (Cross-View) | accuracy %: 89.0 (3.3% higher than the standard LSTM) | Improved |
| increase NRNM block size beyond 8 (ablation) | NRNM · NTU RGB+D (Cross-View) | accuracy % | Hurt |
| integrate NRNM at multiple layers instead of once (ablation) | NRNM on 3-layer LSTM · NTU RGB+D (Cross-View) | accuracy % | Hurt |
| integrate NRNM at the middle layer instead of other layers (ablation) | NRNM on 3-layer LSTM · NTU RGB+D (Cross-View) | accuracy % | Improved |
| use sliding window size around 4 to 8 for memory updates (ablation) | NRNM · NTU RGB+D (Cross-View) | accuracy % | Improved |
| add more layers or hidden units to LSTM instead of memory (ablation) | LSTM · NTU RGB+D (Cross-View) | accuracy %: 84.0 -> 83.1 | No clear effect |
| add more layers to EleAtt-LSTM (ablation) | EleAtt-LSTM · NTU RGB+D (Cross-View) | accuracy %: 85.5 -> 82.7 | Hurt |
| use high-order connections in RNN instead of vanilla RNN (ablation) | vanilla RNN · NTU RGB+D / IMDB | accuracy % | Improved |
Closest papers by meaning
- Learning Sequence Representations by Non-local Recurrent Neural Memory (IJCV 2022 · similarity 0.95)
- Towards Non-Saturating Recurrent Units for Modelling Long-Term Dependencies (AAAI 2019 · similarity 0.87)
- State-Frequency Memory Recurrent Neural Networks (ICML 2017 · similarity 0.87)
- Long Expressive Memory for Sequence Modeling (ICLR 2022 · similarity 0.87)
- Kernel-Based Approaches for Sequence Modeling: Connections to Neural Methods (NeurIPS 2019 · similarity 0.86)
- Temporal Attention-Gated Model for Robust Sequence Classification (CVPR 2017 · 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.
