Research map / Large language models / Language, safety and interpretability
Learning to Parse and Reconstruct: Bidirectional Modeling of Question-to-Program Mapping
Zeying Duan, Youtian Du, Yuanlin Chang, Bowen Lin, Weijia Wu
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 question reconstruction auxiliary loss to program parsing | Transformer program parser · CLEVR val (10K train) | sequence-level accuracy: 0.093 -> 0.691 | Improved |
| add question reconstruction auxiliary loss | Transformer program parser · GQA subset val (30K train) | sequence-level accuracy: 0.517 -> 0.771 | Improved |
| add question reconstruction with large training set (700K) | Transformer program parser · CLEVR val (700K train) | sequence-level accuracy: 1.000 -> 1.000 | No clear effect |
| add question reconstruction auxiliary loss | Transformer program parser · GQA subset val (700K train) | sequence-level accuracy: 0.818 -> 0.833 | Improved |
| add question reconstruction auxiliary loss | Transformer program parser · CLEVR val (30K train) | convergence speed: BiPaR reaches about 0.9 accuracy within 30 epochs; Transformer only at end of training | Improved |
| set loss weight lambda to 0.7 (vs smaller or larger) (ablation) | BiPaR-Full · CLEVR val (10K, 30K, 50K train) | sequence-level accuracy | Improved |
| use decoder-only instead of encoder-decoder reconstruction module (ablation) | BiPaR · CLEVR and GQA subset | sequence-level accuracy: CLEVR 10K: 0.676 -> 0.691 | No clear effect |
| add 50K unlabeled questions with pseudo programs to 10K labeled | BiPaR-Full · CLEVR val | sequence-level accuracy: 0.676 -> 0.932 | Improved |
| add 50K unlabeled questions with pseudo programs to 10K labeled | BiPaR-Full · GQA subset val | sequence-level accuracy: 0.579 -> 0.671 | Improved |
| increase unlabeled questions from 5K to 50K (ablation) | BiPaR-Full · CLEVR val | sequence-level accuracy: 0.876 -> 0.932 | Improved |
Closest papers by meaning
- Neural Semantic Parsing in Low-Resource Settings with Back-Translation and Meta-Learning (AAAI 2020 · similarity 0.85)
- The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervision (ICLR 2019 · similarity 0.85)
- Understanding Information Storage and Transfer in Multi-Modal Large Language Models (NeurIPS 2024 · similarity 0.84)
- See Less, See Right: Bi-directional Perceptual Shaping For Multimodal Reasoning (CVPR 2026 · similarity 0.84)
- Transformers Provably Solve Parity Efficiently with Chain of Thought (ICLR 2025 · similarity 0.84)
- A General Framework for Producing Interpretable Semantic Text Embeddings (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.
