Research map / Architectures and efficiency / Attention and transformers
Systematic Generalization with Edge Transformers
Leon Bergen, Timothy J. O'Donnell, Dzmitry Bahdanau
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 |
|---|---|---|---|
| value ablation: use only edge (i,l) for triangle values (ablation) | Edge Transformer · CLUTRR train k=2,3, test k=6 | accuracy: 82.0 -> 54.4 | Hurt |
| value ablation (ablation) | Edge Transformer · CLUTRR train k=2,3,4, test k=10 | accuracy: 92.1 -> 75.2 | Hurt |
| value ablation (ablation) | Edge Transformer · CFQ dependency parsing MCD3 | EM: 89.1 -> 84.5 | Hurt |
| value ablation (ablation) | Edge Transformer · COGS | generalization accuracy: 87.4 -> 87.8 | No clear effect |
| attention ablation: key from x_ij instead of x_lj (ablation) | Edge Transformer · CLUTRR train k=2,3,4, test k=10 | accuracy: 92.1 -> 78.7 | Hurt |
| attention ablation (ablation) | Edge Transformer · COGS | generalization accuracy: 87.4 -> 86.3 | Hurt |
| attention ablation (ablation) | Edge Transformer · CFQ dependency parsing MCD3 | EM: 89.1 -> 88.1 | No clear effect |
| attention ablation (ablation) | Edge Transformer · CLUTRR train k=2,3, test k=6 | accuracy: 82.0 -> 84.4 | Mixed |
| untied weights across layers (ablation) | Edge Transformer · CFQ dependency parsing MCD3 | EM: 89.1 -> 37.2 | Hurt |
| untied weights across layers (ablation) | Edge Transformer · CLUTRR train k=2,3, test k=10 | accuracy: 58.1 -> 55.3 | Hurt |
| untied weights across layers (ablation) | Edge Transformer · CLUTRR train k=2,3,4, test k=6 | accuracy: 98.1 -> 98.2 | No clear effect |
| untied weights across layers (ablation) | Edge Transformer · COGS | generalization accuracy: 87.4 -> 86.3 | No clear effect |
Closest papers by meaning
- When can transformers reason with abstract symbols? (ICLR 2024 · similarity 0.87)
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.86)
- From Sequence to Structure: Uncovering Substructure Reasoning in Transformers (NeurIPS 2025 · similarity 0.86)
- Grokking of Implicit Reasoning in Transformers: A Mechanistic Journey to the Edge of Generalization (NeurIPS 2024 · similarity 0.86)
- Extrapolation by Association: Length Generalization Transfer In Transformers (NeurIPS 2025 · similarity 0.86)
- The Structure of Relation Decoding Linear Operators in Large Language Models (NeurIPS 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.
