深度学习依赖于一种非常特殊的神经网络:那些叠加了几个神经层的网络。近几年来,深度学习在图像分析、语音识别、自然语言处理等领域取得了重大突破,但目前还没有理论解释。特别是,目前尚不清楚为什么网络越深,它实际上表现得越好。我们认为,这种解释与从我们周围的宇宙中收集的数据的一个关键特征密切相关,这些数据是为了满足机器学习算法的需要:大的不可并行的逻辑深度。粗略地说,我们推测,宇宙的最短计算描述是具有固有的大计算时间的算法,即使有大量的计算机可用于并行化。有趣的是,这个猜想,结合理论计算机科学中的民间传说猜想$ P \neq NC$,解释了深度学习的成功。
英文摘要:
Deep learning relies on a very specific kind of neural networks: those superposing several neural layers. In the last few years, deep learning achieved major breakthroughs in many tasks such as image analysis, speech recognition, natural language processing, and so on. Yet, there is no theoretical explanation of this success. In particular, it is not clear why the deeper the network, the better it actually performs.
We argue that the explanation is intimately connected to a key feature of the data collected from our surrounding universe to feed the machine learning algorithms: large non-parallelizable logical depth. Roughly speaking, we conjecture that the shortest computational descriptions of the universe are algorithms with inherently large computation times, even when a large number of computers are available for parallelization. Interestingly, this conjecture, combined with the folklore conjecture in theoretical computer science that $ P \neq NC$, explains the success of deep learning.
PATCH: a deep learning method to assess heterogeneity of artistic practice in historical paintings
Case Western Reserve University(凯斯西储大学);Department of Physics(物理系);Department of Art History and Art(艺术史与艺术系);The Frick Pittsburgh Museums and Gardens(弗里克匹兹堡博物馆和花园);Department of Computer and Data Sciences(计算机与数据科学系);MORE Center(MORE中心);Cleveland Institute of Art(克利夫兰艺术学院);Factum Foundation(Factum基金会)
ZHAW Datalab & School of Engineering(苏黎世应用科技大学数据实验室与工程学院);Ulm University(乌尔姆大学);ARGUS DATA INSIGHTS Schweiz AG(ARGUS DATA INSIGHTS瑞士股份公司);Deep Impact AG(Deep Impact股份公司);Ca’ Foscari University of Venice(威尼斯福斯卡里大学);Université de Neuchâtel(纳沙泰尔大学);PricewaterhouseCoopers AG(普华永道股份公司);IDSIA Dalle Molle Institute for Artificial Intelligence(IDSIA达莱莫莱人工智能研究所)
From Principles to Practice: A Deep Dive into AI Ethics and Regulations
University of New South Wales(新南威尔士大学);University of Newcastle(纽卡斯尔大学);Swinburne University of Technology(斯威本科技大学);Commonwealth Scientific and Industrial Research Organisation (CSIRO)(联邦科学与工业研究组织(CSIRO))
The 2025 OpenAI Preparedness Framework does not guarantee any AI risk mitigation practices: a proof-of-concept for affordance analyses of AI safety policies
Agrifood Innovation Institute, Australian National University(澳大利亚国立大学农业食品创新研究院);MIT FutureTech, Massachusetts Institute of Technology(麻省理工学院未来科技研究所);The University of Queensland(昆士兰大学);School of Cybernetics, Australian National University(澳大利亚国立大学控制论学院);School of Sociology, Australian National University(澳大利亚国立大学社会学学院);Department of Sociology, Vanderbilt University(范德堡大学社会学系);Center for Democracy and Technology(民主与技术中心)
本文基于可供性理论分析 OpenAI 的 Preparedness Framework,发现该政策未能有效保证风险缓解,表明 AI 安全需超越行业自律的更稳健治理干预。