Research map / Architectures and efficiency / Attention and transformers
Span-Based Neural Buffer: Towards Efficient and Effective Utilization of Long-Distance Context for Neural Sequence Models
Yangming Li, Kaisheng Yao, Libo Qin, Shuang Peng, Yijia Liu, Xiaolong Li
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 span-based neural buffer (SNB) | AWD-LSTM · PTB test | perplexity: 57.30 -> 54.92 | Improved |
| add SNB | AWD-LSTM · WikiText-2 test | perplexity: 65.80 -> 60.43 | Improved |
| add SNB | AWD-LSTM + MoS · PTB test | perplexity: 54.44 -> 52.62 | Improved |
| add SNB | AWD-LSTM + MoS + PS + AT · PTB test (dynamic evaluation) | perplexity: 46.01 -> 43.87 | Improved |
| add SNB | AWD-LSTM + MoS + PS + AT · WikiText-2 test (dynamic evaluation) | perplexity: 38.07 -> 35.16 | Improved |
| add SNB | AWD-LSTM + MoS + PS · WikiText-2 test | perplexity: 59.98 -> 57.87 | Improved |
| remove temperature annealing in training (ablation) | AWD-LSTM+MoS+PS+AT+SNB · PTB test | perplexity: 43.87 -> 45.64 | Hurt |
| remove temperature annealing in training (ablation) | AWD-LSTM+MoS+PS+AT+SNB · WikiText-2 test | perplexity: 35.16 -> 38.23 | Hurt |
| remove temperature annealing at test time (ablation) | AWD-LSTM+MoS+PS+AT+SNB · PTB test | perplexity: 43.87 -> 44.12 | Hurt |
| remove temperature annealing at test time (ablation) | AWD-LSTM+MoS+PS+AT+SNB · WikiText-2 test | perplexity: 35.16 -> 37.45 | Hurt |
| remove intrinsic reward term (ablation) | AWD-LSTM+MoS+PS+AT+SNB · PTB test | perplexity: 43.87 -> 45.92 | Hurt |
| remove intrinsic reward term (ablation) | AWD-LSTM+MoS+PS+AT+SNB · WikiText-2 test | perplexity: 35.16 -> 38.76 | Hurt |
Closest papers by meaning
- The Missing Alignment Link of In-context Learning on Sequences (ICML 2025 · similarity 0.86)
- Proteus: Incremental Memory Activation for Long-Context Sequence Modeling (NeurIPS 2026 · similarity 0.86)
- Understanding and Improving Length Generalization in Recurrent Models (ICML 2025 · similarity 0.85)
- Short-Context Dominance: How Much Local Context Natural Language Actually Needs? (NeurIPS 2026 · similarity 0.85)
- Sequence to Sequence Reward Modeling: Improving RLHF by Language Feedback (AAAI 2025 · similarity 0.85)
- InfLLM: Training-Free Long-Context Extrapolation for LLMs with an Efficient Context Memory (NeurIPS 2024 · 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.
