Research map / Architectures and efficiency / Attention and transformers
Not All Attention Is Needed: Gated Attention Network for Sequence Data
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 global soft attention with gated sparse attention | BiLSTM text classifier · IMDB | accuracy: 0.8863 -> 0.8941 | Helped |
| replace global soft attention with gated sparse attention | BiLSTM text classifier · AG's News | accuracy: 0.9264 -> 0.9263 | Hurt |
| replace global soft attention with gated sparse attention | BiLSTM text classifier · SST-1 | accuracy: 0.4445 -> 0.4464 | Helped |
| replace global soft attention with gated sparse attention | BiLSTM text classifier · SST-2 | accuracy: 0.8246 -> 0.8262 | Helped |
| replace global soft attention with gated sparse attention | BiLSTM text classifier · TREC | accuracy: 0.9008 -> 0.9124 | Helped |
| use dynamically gated sparse attention instead of global soft attention | BiLSTM text classifier · IMDB | attention FLOPs: 2.4G -> 0.4G | Helped |
| use dynamically gated sparse attention instead of global soft attention | BiLSTM text classifier · AG's News | attention FLOPs: 131M -> 59M | Helped |
| use dynamically gated sparse attention instead of global soft attention | BiLSTM text classifier · SST-1 | attention FLOPs: 17M -> 6M | Helped |
| use dynamically gated sparse attention instead of global soft attention | BiLSTM text classifier · SST-2 | attention FLOPs: 14.1M -> 5.6M | Helped |
| use dynamically gated sparse attention instead of global soft attention | BiLSTM text classifier · TREC | attention FLOPs: 1.4M -> 0.7M | Helped |
| replace the LSTM auxiliary network with a feed-forward network (ablation) | GA-Net · IMDB | accuracy: 0.8941 -> 0.8890 | Hurt |
| replace the LSTM auxiliary network with a self-attention network (ablation) | GA-Net · IMDB | accuracy: 0.8941 -> 0.8892 | Hurt |
Closest papers by meaning
- Temporal Attention-Gated Model for Robust Sequence Classification (CVPR 2017 · similarity 0.89)
- Word Attention for Sequence to Sequence Text Understanding (AAAI 2018 · similarity 0.88)
- Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free (NeurIPS 2025 · similarity 0.86)
- Asymptotics of SGD in Sequence-Single Index Models and Single-Layer Attention Networks (NeurIPS 2025 · similarity 0.85)
- Fundamental limits of learning in sequence multi-index models and deep attention networks: high-dimensional asymptotics and sharp thresholds (ICML 2025 · similarity 0.84)
- Explaining Modern Gated-Linear RNNs via a Unified Implicit Attention Formulation (ICLR 2025 · 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.