Autonomous Research for Open-Ended Problems: A Case Study on Telecom Ticket Retrieval
Source: https://arxiv.org/abs/2609.13073
Author: Junghyun Min, Huseyin Uzunalioglu, Mohamed Trabelsi
Published: 2026-09-11
Platform: arxiv
中文概要
arXiv 2609.13073（cs.AI）研究如何将全自动 ML 研究从窄搜索空间（语言建模、生物医学基准）扩展到开放式工业级问题，以电信工单检索为案例，涉及表征、架构与训练数据生成等多自由度。对比商业与开源自主研究智能体在该案例上的表现与局限，给出哪些维度先失效的经验证据。
English Summary
arXiv 2609.13073, cs.AI. Studies how fully-autonomous ML research can be extended beyond narrow search spaces (LM, biomed benchmarks) to open-ended industry-grade problems, using telecom ticket retrieval as a case study with degrees of freedom in representation, architecture, and training-data generation. Reports that commercial and open-source autonomous-research agents show both promise and limitations on the case study; details on which dimensions break first.