Research map / Architectures and efficiency / Attention and transformers
Symmetry-Aware Transformer Training for Automated Planning
Markus Fritzsche, Elliot Gestrin, Jendrik Seipp
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 contrastive object-name equivariance loss (ablation) | SymT-E and SymT-ED · Blocksworld, Gripper, Visitall, Logistics | training divergence: 1 diverged with loss vs 19 without | Improved |
| add contrastive loss (ablation) | SymT-E (heuristic model) · four planning domains | coverage | Improved |
| add contrastive loss (ablation) | SymT-ED (plan generation) · four planning domains | validation loss and coverage | Mixed |
| use SymT-ED instead of PlanGPT for extrapolation | plan generation · extrapolation sets | coverage | Improved |
| use SymT-ED instead of encoder-only SymT-E | plan generation · four planning domains | coverage | Improved |
| use SymT instead of PlanGPT in Logistics | plan generation · Logistics | coverage | Hurt |
| rename Visitall locations randomly (ablation) | PlanGPT · Visitall (original data) | coverage [%]: 93.80 -> 0 | Hurt |
| rename Floortile robots randomly (ablation) | PlanGPT · Floortile (original data) | coverage [%]: 44.73 -> 0.05 | Hurt |
| rename Blocksworld blocks randomly (ablation) | PlanGPT · Blocksworld (original data) | coverage [%]: 99.55 -> 99.36 | No clear effect |
| use regrounding vs applicability-filtered generation (ablation) | SymT models · extrapolation | coverage | Mixed |
Closest papers by meaning
- SiT: Symmetry-invariant Transformers for Generalisation in Reinforcement Learning (ICML 2024 · similarity 0.86)
- Transformers Provably Learn Chain-of-Thought Reasoning with Length Generalization (NeurIPS 2025 · similarity 0.85)
- From Shortcut to Induction Head: How Data Diversity Shapes Algorithm Selection in Transformers (NeurIPS 2025 · similarity 0.84)
- Multi-head Transformers Provably Learn Symbolic Multi-step Reasoning via Gradient Descent (NeurIPS 2025 · similarity 0.84)
- Towards Understanding Inductive Bias in Transformers: A View From Infinity (ICML 2024 · similarity 0.84)
- Do Efficient Transformers Really Save Computation? (ICML 2024 · 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.
