近日，机器之心AIxiv专栏发布了一篇关于ICML 2024会议的高分论文。该研究围绕“零阶优化器微调大模型”展开，可大幅降低内存需求，为开源大语言模型（LLM）的广泛应用开辟了新的道路。
论文共同第一作者包括张逸骅、李平治、洪骏远和李佳翔等多位年轻学者。他们分别来自密歇根州立大学、北卡罗来纳大学教堂山分校、德州大学奥斯汀分校等顶尖学府，研究领域涵盖大模型的安全、隐私和效率问题，以及高效机器学习和AI4Science等领域。
随着开源大语言模型（LLM）的兴起，如何使这些模型适应各种下游任务成为研究热点。微调（fine-tuning）是最广泛采用的方法。而该论文提出的零阶优化器微调大模型，能有效降低内存需求，为LLM在更多领域的应用提供了可能。
据了解，该论文的研究工作得到了导师们的高度认可，并在学术界引起了广泛关注。机器之心AIxiv专栏作为发布学术、技术内容的栏目，过去数年接收报道了2000多篇内容，覆盖全球各大高校与企业的顶级实验室，有效促进了学术交流与传播。
此外，机器之心欢迎广大研究人员积极投稿分享优秀工作。投稿邮箱为：liyazhou@jiqizhixin.com和zhaoyunfeng@jiqizhixin.com。随着研究的深入，期待更多创新成果涌现，推动人工智能领域的发展。
英语如下：
News Title: “ICML 2024 Amazing Paper Unveils: Zero-Order Optimizer Fine-Tunning Big Model, Revolutionizing Open Source LLM Technology”
Keywords: ICML high-scoring paper, fine-tuning big model, open source LLM
News Content:
ICML 2024 Front-Edge Paper Unveils: Zero-Order Optimizer Fine-Tuning Big Model to Significantly Reduce Memory Requirements
Recently, the “Machine Intelligence” AIxiv column published a highly scored paper from the ICML 2024 conference. The research focuses on “zero-order optimizer fine-tuning big model,” which can significantly reduce memory requirements, opening new paths for the widespread application of open-source large language models (LLMs).
The paper’s co-first authors include young scholars such as Zhang Yihua, Li Pingzhi, Hong Junyuan, and Li Jiaxiang. They come from top universities like Michigan State University, University of North Carolina at Chapel Hill, and University of Texas at Austin, and their research fields cover the safety, privacy, and efficiency of large models, as well as efficient machine learning and AI4Science.
With the rise of open-source large language models (LLMs), how to adapt these models to various downstream tasks has become a research hotspot. Fine-tuning is the most widely used method. The paper’s proposed zero-order optimizer fine-tuning big model effectively reduces memory requirements, making it possible for LLMs to be applied in more fields.
It is understood that the paper’s research work has been highly recognized by mentors and has attracted widespread attention in the academic community. As a column for publishing academic and technical content, the Machine Intelligence AIxiv column has received over 2,000 reports in the past few years, covering top laboratories from major universities and enterprises worldwide, effectively promoting academic exchange and dissemination.
In addition, Machine Intelligence welcomes active contributions from researchers to share excellent work. The submission email addresses are [liyazhou@jiqizhixin.com](mailto:liyazhou@jiqizhixin.com) and [zhaoyunfeng@jiqizhixin.com](mailto:zhaoyunfeng@jiqizhixin.com). With ongoing research, we look forward to more innovative achievements to promote the development of the artificial intelligence field.
【来源】https://www.jiqizhixin.com/articles/2024-07-04-8
Views: 1