The field of machine learning for scientific discovery is rapidly evolving, with a focus on developing innovative methods for equation discovery, automated mathematical theory formation, and knowledge-informed feature extraction. Recent research has highlighted the importance of bridging symbolic reasoning with geometric reconstruction, enabling principled benchmarking of progress in compositional generalization and data-driven scientific induction. Notable developments include the introduction of comprehensive benchmarks for symbolic surface discovery and the evaluation of autonomous AI scientists in radiation biology.

Noteworthy papers include: SurfaceBench, which establishes a challenging testbed for evaluating equation discovery quality and generalization across representation types and surface complexities. Rogue One, a novel LLM-based multi-agent framework for knowledge-informed automatic feature extraction, which significantly outperforms state-of-the-art methods on a comprehensive suite of datasets.