Research map / Architectures and efficiency / Attention and transformers
Understanding How Encoder-Decoder Architectures Attend
Kyle Aitken, Vinay Venkatesh Ramasesh, Yuan Cao, Niru Maheswaranathan
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 two input-dependent terms to the temporal-only alignment approximation | alignment score approximation · eSCAN | alignment variance captured (%): >77% -> >87% | Improved |
| subtract temporal component from hidden states (AED) | encoder hidden states · eSCAN | word-state variance drop (%): 90% | Improved |
| subtract temporal component from hidden states (AED) | encoder hidden states · English-French translation | word-state variance drop (%): 8% | No clear effect |
| subtract temporal component from hidden states (AO) | encoder hidden states · English-French translation | word-state variance drop (%): 25% | No clear effect |
| swap GRU cell for LSTM in AED | AED · one-to-one task | attention pattern | Mixed |
| swap GRU for UGRNN or LSTM cells | AED/AO · one-to-one task | test accuracy (%): 100% | No clear effect |
Closest papers by meaning
- Rethinking Attention: Exploring Shallow Feed-Forward Neural Networks as an Alternative to Attention Layers in Transformers (Student Abstract) (AAAI 2024 · similarity 0.87)
- Attention Alignment Between Humans and Vision-Language Models (NeurIPS 2026 · similarity 0.86)
- Forgetting Transformer: Softmax Attention with a Forget Gate (ICLR 2025 · similarity 0.86)
- What are you sinking? A geometric approach on attention sink (NeurIPS 2025 · similarity 0.86)
- Finding Interpretable Prompt-Specific Circuits in Language Models (NeurIPS 2026 · similarity 0.86)
- Mind the Gap: a Spectral Analysis of Rank Collapse and Signal Propagation in Attention Layers (ICML 2025 · 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.
