Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Pipelines with Limited Data and Domain Knowledge: A Study in Parsing Physics Problems
Mrinmaya Sachan, Kumar Avinava Dubey, Tom M. Mitchell, Dan Roth, Eric P Xing
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 |
|---|---|---|---|
| learn the pipeline jointly with trust model instead of traditional pipeline with AdaBoost combination (ablation) | Nuts&Bolts · Physics diagram parsing | F1: 0.66 -> 0.74 | Improved |
| add trust modelling of function approximators (ablation) | Nuts&Bolts · Physics diagram parsing | F1: 0.64 -> 0.74 | Improved |
| learn rule weights instead of fixing them to 1 (ablation) | Nuts&Bolts · Physics diagram parsing | F1: 0.59 -> 0.74 | Improved |
| add all unlabelled diagrams to semi-supervised training (ablation) | Nuts&Bolts · Physics diagram parsing | F1: 0.58 -> 0.74 | Improved |
| reduce text-element supervision from 207 to 25 diagrams (ablation) | Nuts&Bolts · Physics diagram parsing | F1: 0.74 -> 0.61 | Hurt |
| Nuts&Bolts diagram parser instead of EdgeBoxes/G-ALIGNER/DSDP-Net baselines | Nuts&Bolts · Physics diagram parsing | F1: 0.65 -> 0.74 | Improved |
| use Nuts&Bolts diagram parses in question answering | Parsing to Programs solver · Textbook physics QA | accuracy: 23 -> 32 | Improved |
| Nuts&Bolts instead of Snorkel for relation extraction | Nuts&Bolts · CDR relation extraction | F1: 45.3 -> 47.1 | Improved |
| Nuts&Bolts instead of Snorkel for relation extraction | Nuts&Bolts · Genomics relation extraction | F1: 57.2 -> 57.3 | No clear effect |
| Nuts&Bolts semi-supervised instead of semi-supervised PCFG | Nuts&Bolts · Physics diagram parsing | F1: 0.58 -> 0.74 | Improved |
Closest papers by meaning
- Scaling Physical Reasoning with the PHYSICS Dataset (NeurIPS 2025 · similarity 0.86)
- UGPhysics: A Comprehensive Benchmark for Undergraduate Physics Reasoning with Large Language Models (ICML 2025 · similarity 0.86)
- SciCode: A Research Coding Benchmark Curated by Scientists (NeurIPS 2024 · similarity 0.84)
- SciInstruct: a Self-Reflective Instruction Annotated Dataset for Training Scientific Language Models (NeurIPS 2024 · similarity 0.84)
- LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models (ICML 2025 · similarity 0.84)
- ConFIG: Towards Conflict-free Training of Physics Informed Neural Networks (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.
