Research map / Architectures and efficiency / Attention and transformers
Novel positional encodings to enable tree-based transformers
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 sinusoidal positional encodings with tree positional encodings (seq-to-tree) | Transformer · ATIS | whole program accuracy (%): 84.4 -> 86.4 | Helped |
| replace sinusoidal positional encodings with tree positional encodings (seq-to-tree) | Transformer · GEO | whole program accuracy (%): 81.1 -> 84.6 | Helped |
| replace sinusoidal positional encodings with tree positional encodings (seq-to-tree) | Transformer · JOBS | whole program accuracy (%): 85.0 -> 84.3 | Hurt |
| tree-transformer vs sequence-transformer on long synthetic programs | Transformer · Synthetic long-program translation | whole program error rate | Helped |
| depth-first instead of breadth-first decoding traversal (ablation) | Tree-transformer · Synthetic translation (short and long) | whole program error rate | Helped |
| tree-transformer instead of tree2tree LSTM | Tree-transformer · CoffeeScript-JavaScript translation | whole program error rate: up to 20 percentage point gain on hardest tasks | Mixed |
| tree-transformer instead of tree2tree LSTM | Tree-transformer · Synthetic translation | whole program error rate | Helped |
| seq-to-tree transformer vs best recurrent literature | Seq2Tree Transformer · ATIS | whole program accuracy (%): 84.6 -> 86.4 | Helped |
Closest papers by meaning
- How Transformers Learn Structured Data: Insights From Hierarchical Filtering (ICML 2025 · similarity 0.86)
- Integrating Tree Path in Transformer for Code Representation (NeurIPS 2021 · similarity 0.86)
- PaTH Attention: Position Encoding via Accumulating Householder Transformations (NeurIPS 2025 · similarity 0.85)
- The Power of Hard Attention Transformers on Data Sequences: A formal language theoretic perspective (NeurIPS 2024 · similarity 0.84)
- Algebraic Positional Encodings (NeurIPS 2024 · similarity 0.84)
- Round and Round We Go! What makes Rotary Positional Encodings useful? (ICLR 2025 · similarity 0.83)
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.