arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2607.08077cs.LG
收藏

模块化预训练实现访问控制

Modular Pretraining Enables Access Control

AE Studio(AE工作室);Anthropic

Ethan Roland, Murat Cubuktepe, Erick Martinez, Stijn Servaes, Keenan Pepper, Mike Vaiana, Diogo Schwerz de Lucena, Judd Rosenblatt, Addie Foote, Cem Anil, Alex Cloud

AI总结:

针对人工智能两用困境及多模型部署成本高问题,提出梯度路由辅助模块(GRAM)预训练方法,能在推理时消融模块功能。实验显示其可有效禁用目标功能且抗微调恢复,训练成本低,紧密跟踪数据过滤。

摘要:

人工智能开发者面临两用困境,一种人工智能能力既能帮助用户治病,也能被用于合成有害物质。访问控制可解决此问题,但训练和部署多个模型成本过高。为此提出梯度路由辅助模块(GRAM),一种预训练方法,通过添加模块并选择性更新来诱导专业化。在推理时消融模块可去除其功能。实验表明,GRAM能禁用目标功能并保留其他功能,比事后遗忘更能抵抗微调恢复。Chinchilla最优缩放分析表明,数据过滤模型与全数据模型在去除功能上的差距随规模扩大,在保留功能上差距小,GRAM紧密跟踪数据过滤。GRAM训练成本与支持的能力配置文件数量无关,在五配置设置中比数据过滤成本降低5倍。

英文摘要:

AI developers face a dual-use dilemma. An AI capability that helps one user cure a disease can help another synthesize one. This dilemma could be resolved with access control, limiting dual-use AI capabilities to trusted deployments with a legitimate need. A gold standard for access control would be to serve separate models with different capabilities to different users. However, training and deploying multiple models is prohibitively expensive. To address this challenge, we propose gradient-routed auxiliary modules (GRAM), a pre-training method that adds modules to a neural network and selectively updates them to induce specialization. Ablating a module at inference time removes its capability from the network, approximating a model trained on filtered data. We evaluate GRAM on synthetic stories and realistic dual-use data spanning virology, cybersecurity, nuclear physics, and specialized code. These experiments show that GRAM disables targeted capabilities while preserving the rest, and resists their recovery under finetuning better than post-hoc unlearning. Most importantly, a Chinchilla-optimal scaling analysis from 50M to 5B parameters shows that the gap between data-filtered and full-data models widens with scale on removed capabilities but stays small on retained ones, and that GRAM closely tracks data filtering. GRAM's training cost is independent of the number of supported capability profiles, yielding a 5x reduction over data filtering in our 5-profile setting.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑