Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.29014.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

AlphaDiverse is a framework for post-training local LLM-based multi-agent systems for alpha factor mining. It uses diverse research path collection, supervised fine-tuning, and a joint GRPO method to address research path collapse, with experiments on four Chinese stock universes.

Concrete recipe for avoiding research-path collapse in LLM quant agents, with SFT and GRPO details. Useful for quants building local multi-agent research pipelines, narrower outside that niche.

Cited by

No citations on record.