Research map / Architectures and efficiency / RNNs and normalization
Modeling Local Dependence in Natural Language with Multi-Channel 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 MC-RNN multi-channel, multi-block local connections and channel attention | Baseline-RNN on IWSLT 2014 De-En · IWSLT 2014 De-En | BLEU: 31.03 -> 32.23 | Helped |
| add MC-RNN multi-channel, multi-block local connections and channel attention | Baseline-RNN on Gigaword · Gigaword test | ROUGE-1 F1: +1.90 | Helped |
| add MC-RNN multi-channel, multi-block local connections and channel attention | Baseline-RNN on Gigaword · Gigaword test | ROUGE-2 F1: +1.45 | Helped |
| add MC-RNN multi-channel, multi-block local connections and channel attention | Baseline-RNN on Gigaword · Gigaword test | ROUGE-L F1: +1.48 | Helped |
| add MC-RNN local multi-channel connections | AWD-LSTM w/o finetune on Penn Treebank · Penn Treebank validation | perplexity: 60.7 -> 59.2 | Helped |
| add MC-RNN local multi-channel connections | AWD-LSTM w/o finetune on Penn Treebank · Penn Treebank test | perplexity: 58.8 -> 56.9 | Helped |
| use 3 channels rather than 2 (ablation) | MC-RNN on IWSLT 2014 De-En · IWSLT 2014 De-En | BLEU: 31.98 -> 32.23 | Helped |
| use 4 channels rather than 3 (ablation) | MC-RNN on IWSLT 2014 De-En · IWSLT 2014 De-En | BLEU: 32.23 -> 32.09 | Hurt |
| use 3 channels rather than 2 (ablation) | MC-RNN on Gigaword · Gigaword test | ROUGE-1 F1: 36.21 -> 36.55 | Helped |
| use 3 channels rather than 2 (ablation) | MC-RNN on Gigaword · Gigaword test | ROUGE-2 F1: 17.30 -> 17.58 | Helped |
| use 3 channels rather than 2 (ablation) | MC-RNN on Gigaword · Gigaword test | ROUGE-L F1: 33.60 -> 33.72 | Helped |
| use 4 channels rather than 3 (ablation) | MC-RNN on Gigaword · Gigaword test | ROUGE-1 F1: 36.55 -> 36.50 | Hurt |
Closest papers by meaning
- RNN-Based Sequence-Preserved Attention for Dependency Parsing (AAAI 2018 · similarity 0.89)
- Recurrent Hierarchical Topic-Guided RNN for Language Generation (ICML 2020 · similarity 0.88)
- TopicRNN: A Recurrent Neural Network with Long-Range Semantic Dependency (ICLR 2017 · similarity 0.88)
- Learning Hierarchical Structures with Differentiable Nondeterministic Stacks (ICLR 2022 · similarity 0.87)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · similarity 0.86)
- RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling (NeurIPS 2025 · similarity 0.85)
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.