Researchers release MioFFAn, an open-source framework for annotating and formalizing scientific formulas with LLM-assisted workflows
The tool targets a documented scarcity of high-quality datasets for translating mathematical expressions into executable symbolic code, introducing modular automation and human-in-the-loop evaluation.
1 source · cross-referenced
- MioFFAn is an open-source, document-centric framework designed to accelerate annotation for formula formalization, a task hindered by a scarcity of high-quality, domain-specific datasets.
- The framework builds on the MioGatto architecture and adds features such as equation selection and aided symbolic code specification to support diverse scientific fields.
- MioFFAn incorporates partial automation via LLMs through modular sub-tasks with strict output formats, enabling iterative refinement and evaluation using standard NLP metrics.
- Authors report a preliminary evaluation demonstrating the efficacy of the human-in-the-loop approach in improving annotation workflows.
Researchers from Universitat Pompeu Fabra and the International Center for Numerical Methods in Engineering introduced MioFFAn, an open-source framework aimed at addressing a documented scarcity of high-quality, ground-truth datasets for formula formalization—the task of translating mathematical expressions in scientific literature into executable symbolic code.
Built on the MioGatto architecture, MioFFAn extends prior features to overcome structural limitations and introduces domain-specific functionalities such as selection of equations of interest and aided symbolic code specification. The framework allows users to configure custom taxonomies and properties for identified symbols and compatible symbolic operators, making it adaptable to specialized scientific fields.
MioFFAn is designed to integrate partial automation via large language models through a modular set of automated sub-tasks with strict output formats. This design enables researchers to iteratively refine automation strategies and compare approaches using standard NLP metrics.
The authors describe a preliminary evaluation that demonstrates the efficacy of the human-in-the-loop approach, positioning MioFFAn as a practical tool for accelerating annotation workflows in technical domains.
- Oct 13, 2026 · Google DeepMind — Blog
Google DeepMind details safety guardrails for Gemini 4 Argon’s cyberdefense and biology capabilities
Trust76 - Oct 13, 2026 · Google DeepMind — Blog
Google DeepMind unveils Gemini 4 Argon for robotics with 87% score on Embodied Reasoning benchmark
Trust78 - Oct 13, 2026 · arXiv cs.CL
Researchers propose Elastic Latent Recurrence to unify recurrent and non-recurrent language models without inference-time trade-offs
Trust79