New AI framework analyzes ancient oracle bone scripts with Qwen2.5-VL

Researchers have developed OracleAnalyser, a new framework designed to analyze the implicit semantics of oracle bone scripts using multimodal large language models (MLLMs). This framework fine-tunes the Qwen2.5-VL-3B-Instruct model through advanced post-training techniques, including a novel preference optimization algorithm called Stable Focal Preference Optimization (SFPO). The project also introduces new datasets for oracle bone reasoning and preferences, along with a benchmark to evaluate analytical capabilities. Experiments demonstrate that OracleAnalyser achieves superior analytical performance, outperforming larger models despite its smaller parameter count. AI

IMPACT This research advances the application of MLLMs in historical text analysis, potentially enabling deeper insights into ancient cultures and languages.

RANK_REASON The cluster describes a new research paper introducing a novel framework and algorithm for analyzing ancient scripts using LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Hugging Face
• OracleAnalyser
• Qwen2.5-VL-3B-Instruct
• Stable Focal Preference Optimization

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →