Reproduce “Plane Geometry Problem Solving with Multi-modal Reasoning: A Survey” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Seunghyuk Cho, Zhenyue Qin, Yang Liu, Youngbin Choi, Seungbeom Lee, Dongwoo Kim
This survey provides a comprehensive overview of recent advancements in Plane Geometry Problem Solving (PGPS), a field requiring multi-modal reasoning from geometric diagrams and textual descriptions. The paper categorizes existing methods within an encoder-decoder framework and identifies key challenges like visual perception errors (42.3%) and hallucination, while also proposing future research directions.
Key insights
- Most existing PGPS methods follow an encoder-decoder architecture, categorizable by rule-based or neural network encoders, and various output formats like theorem sequences or natural language.
- Visual perception errors are the most frequent type of mistake in current PGPS systems, accounting for 42.3% of errors, followed by reasoning errors at 33.9%.
- Current challenges include hallucination, data leakage in benchmarks, and limited reasoning capabilities, pointing towards needs for improved multi-modal integration and robust evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.