Workflow
力学
icon
Search documents
「流匹配」成ICML 2025超热门主题!网友:都说了学物理的不准转计算机
机器之心· 2025-07-13 04:58
机器之心报道 编辑:笑寒 流体力学融入生成式 AI ,构建了一种非常简洁、优雅的形态。 众所周知,第 42 届国际机器学习大会(ICML)将于 7 月 13 日至 19 日在加拿大温哥华盛大举行。 在生成式 AI 领域,最新的前沿热点已经转向探索更高质量,更稳定,更简洁,更通用的模型形态。 流匹配(Flow Matching)技术 正完美的踩中了每一个热点要素。 自从 FLUX 模型发布后,能够处理多种输入类型的流匹配架构逐渐成为目光焦点。 也因此有学者感慨,在 ICML 2025 的生成相关工作中,流匹配技术几乎 无处不在 。 流匹配技术虽说在生成式 AI 领域是前沿研究,但其核心概念来源于流体力学。 令人惊讶的是,物理领域的有关概念在近些年的确为生成领域的研究提供了很多新方向和新成果。 甚至 薛定谔桥都能用在扩散生成领域 ! 在知乎相关技术解读专栏《 深入解析 Flow Matching 技术》下, 网友怒评:物理学专业的不准转计算机! 本文参考研究者 Floor Eijkelboom 的最新推文,从原理入手,避免繁杂的数学公式,来介绍这一简洁优雅且高效的生成技术。 生成:噪声映射到数据 生成工作是一个逐 ...
张朝阳对话诺贝尔奖得主David Gross:解密“时空涌现”“质量起源”
Guang Zhou Ri Bao· 2025-07-12 13:03
7月11日,搜狐创始人、董事局主席兼首席执行官、物理学博士张朝阳和2004年诺贝尔物理学奖获得 者、2025年基础科学终身成就奖得主、美国国家科学院院士、中国科学院外籍院士戴维·格罗斯(David Gross)展开了一场知识对谈,共同探讨物质世界最基础的构成和物理理论的前沿进展。 据悉,张朝阳与格罗斯教授从自然界四种基本力的差异切入,其间讨论了强相互作用的"渐近自由"特性 ——这一突破性发现正是格罗斯教授获得诺贝尔物理学奖的关键。随着对话向理论前沿延伸,二人又从 量子引力谈到弦理论,探究时空的本质,同时也直面了暗物质探测领域的当代难题。 从量子场论的"反叛者"到渐近自由的"奠基人" 作为诺奖级成果,渐近自由的发现历程是张朝阳本次对话最关心的话题之一。格罗斯教授回忆20世纪60 年代末的"泥沼":加速器每周都能发现新粒子,但没人理解它们的本质。直至1968年,斯坦福直线加速 器中心(SLAC)意外证实质子内部存在自由运动的点状粒子——"夸克",一举逆转局面。但这又引出 了更多的矛盾,如果夸克能够自由运动,它们为何从未单独现身? 在张朝阳的追问下,格罗斯教授承认,他在29岁为此开启了一场豪赌。在长达五年的工作中, ...
AI能否解决黎曼猜想等未知难题?诺奖得主这样说
Di Yi Cai Jing· 2025-07-12 10:01
Core Viewpoint - The current AI models are significantly overestimated, serving primarily as tools rather than independent scientific entities [1][2][5] Group 1: AI and Scientific Discovery - David Gross argues that solving major physical or mathematical problems relies on human intelligence and creativity, with AI acting as a powerful auxiliary tool [2][5] - There is skepticism regarding AI's ability to prove complex conjectures within a five-year timeframe, as highlighted by a bet between Zhang Yaqin and mathematician Shing-Tung Yau [1][2] - Gross expresses dissatisfaction with the current capabilities of AI, noting that early versions of ChatGPT struggled with basic tasks like counting [2] Group 2: Nobel Prize and AI - The 2024 Nobel Prize in Physics awarded to John Hopfield is not attributed to AI achievements, as his work extends physical methods into neuroscience [4][5] - Gross emphasizes that Hopfield's research is a continuation of physics rather than a contribution to AI, reinforcing the distinction between the two fields [5] Group 3: Computational Power and Theoretical Physics - The exponential growth in computational power has significantly advanced theoretical physics, allowing for complex calculations that were previously labor-intensive [5] - Gross reflects on the historical limitations of computational methods in quantum chromodynamics (QCD) and how modern advancements have transformed research capabilities [5] Group 4: Encouragement for Young Researchers - Gross encourages young researchers to enjoy the process of exploration and maintain curiosity, emphasizing that the joy of research lies in the journey of discovery [6]
鱼儿悬停水中为何要消耗更多能量
Ke Ji Ri Bao· 2025-07-08 02:07
更精妙的是,不同鱼类的"平衡策略"各具特色:胸鳍位置靠后的鱼类仿佛拥有更长的"操控杆",能 更高效地调整姿态;体型修长的鱼类则需要付出更多能量对抗水流扰动;而那些圆润可爱的鱼,却凭借 紧凑的体型成为悬停界的"节能高手"。 原标题:鱼儿悬停水中为何要消耗更多能量 悬停时的鱼类代谢率比静止或休息时高出整整两倍。这种高耗能的生存策略,实则是数百万年进化 的结晶。譬如那些擅长高速巡游的鱼类,悬停效率更低些,但看似静止的身体,实则是积蓄力量的弹 簧,能像箭般在水域中疾驰;而那些生活在复杂珊瑚礁的鱼类,进化出了更圆润的体型,牺牲速度换取 悬停时的稳定性,可以静静等待猎物上门——这都是它们用身体写出的生存智慧密码。 在南加州圣克莱门特岛附近,一条加里波第鱼在水中悬停。图片来源:斯克里普斯海洋研究所 你在水族箱里见过这样的画面吗:鱼儿静止悬停在水中,仿佛被施了魔法般一动不动,连水流拂过 鱼鳍都纹丝不动。长久以来,科学家都认为这是鱼类最省力的休息姿势,但最新研究却颠覆了这一认知 ——悬停时的鱼儿,其实正在默默燃烧着两倍于休息状态的能量!这背后,是一场关于平衡、进化与流 体力学精妙博弈的故事。 几乎所有硬骨鱼体内都藏着一个神奇的 ...
7月5日日本将有大灾难?预言未来与末日公式
Hu Xiu· 2025-07-04 01:27
Group 1 - The article explores various methods humans have used to predict the future, ranging from ancient divination to modern scientific models, highlighting the curiosity and desire for knowledge about the unknown future [1] - The story of Japanese manga artist Tatsuki Ryo, who gained fame for her supposed precognitive dreams, illustrates the intersection of science and mysticism in predicting future events [2][3] - Ryo's manga "What I Saw in the Future" gained attention after it seemingly predicted the 2011 Tōhoku earthquake, leading to a surge in demand for the book [2][3] - The upcoming date of July 5, 2025, predicted by Ryo as a potential disaster date, has caused significant social impact, including a decline in flight bookings to Japan and increased sales of emergency supplies [4] Group 2 - The article discusses Samuel Benner's predictions based on historical price cycles, which he believed were influenced by natural phenomena like solar activity [15][19] - Benner's work has been both praised for its accuracy and criticized for its simplicity and potential biases, reflecting the ongoing debate about the validity of cyclical predictions in economics [19][20] - The article also mentions John Richard Gott III's doomsday formula, which predicts the potential extinction of humanity within a specific timeframe, showcasing the controversial nature of such predictions [22][23] Group 3 - The development of computer models in the late 20th century aimed to predict societal trends based on various factors, including population growth and resource depletion [31][32] - The "Limits to Growth" report, which utilized these models, faced criticism despite its popularity, while newer models suggest a more optimistic outlook for humanity's future [34][35] - The article concludes that while predictions about the future can be fraught with uncertainty, they serve as a means to better understand the present and prepare for potential challenges [31][34]
不同脑细胞“演员”拿的是统一“剧本”
Ke Ji Ri Bao· 2025-07-01 23:14
Group 1 - The core finding of the research reveals how the brain makes unified decisions through seemingly chaotic yet orderly neural activities, guided by a hidden "script" [1][2] - The study involved training monkeys to judge color dominance, recording neural activity, and found that decision-making is influenced by preference adjustments and neural dynamics [2] - The research indicates that while individual neurons exhibit unique responses, they collectively contribute to a stable decision-making state, akin to actors following a script [2][4] Group 2 - The implications of this discovery extend to understanding decision-making processes and may provide new insights for treating disorders like schizophrenia and bipolar disorder [4] - Future research plans include exploring the connections between different types of neurons and their specific roles in the decision-making process [4]
超清太阳表面图像揭开“磁帘”秘密
Ke Ji Ri Bao· 2025-06-25 23:30
据美国每日科学网站近日报道,美国国家太阳天文台研究团队利用丹尼尔·井上太阳望远镜此前捕获的 超清晰太阳表面图像,完整呈现了太阳表面超精细磁条状特征——"条纹结构"。 这些"条纹结构"将重塑人们对太阳表面微观尺度磁场动力学的认知,有助于更精确预测太阳耀斑、日冕 物质抛射等影响地球的空间天气事件。 山东大学空间科学与技术学院教授郑瑞生解释道:"太阳表面(即光球层)布满了直径约1000公里的对 流元胞——米粒组织,新发现的'条纹结构'就像悬挂于米粒组织边界的'磁帘'。" 这些"磁帘"实际上是太阳表面磁场的帘状电流片结构,如同随风摆动的织物般波动起伏。当高温米粒组 织边界发出的光线穿过这些"磁帘"时,相互作用会产生明暗交替的条纹模式,这可精确反映底层磁场的 空间变化规律。当"磁帘"区域的磁场弱于周边环境时呈现暗条纹,反之则呈现亮条纹。 这项突破性观测得益于井上太阳望远镜可见光宽带成像仪(VBI)在G波段的超高分辨率观测。该波段 特别适合研究太阳,能显著增强磁活动区特征,可更精准地捕捉太阳黑子及更多精细结构。 为解析最新观测结果,研究团队将图像与模拟太阳表面物理过程的前沿数值模型进行了系统性比较。结 果证实,这些"条纹 ...
假想粒子“轴子”,终于被找到了?
Hu Xiu· 2025-06-20 00:38
Group 1 - The axion is a hypothetical particle proposed to solve the strong CP problem in quantum chromodynamics, introduced by theorists Frank Wilczek and Steven Weinberg in 1978 [1][3] - Axions are extremely difficult to detect due to their weak interaction with other particles, yet they are considered a significant component of dark matter in the universe [4][6] - Recent advancements suggest that axions can exist in a quasi-particle form within solid-state materials, referred to as "axion insulators," which exhibit unique magnetic and electric properties [7][8] Group 2 - The material MnBi2Te4 has been predicted to be an axion insulator, featuring a layered structure with alternating magnetic properties, which may allow for the observation of axion quasi-particles [9][11] - Experimental results published in April 2025 demonstrated periodic oscillations of the magnetic-electric coupling coefficient in MnBi2Te4, indicating the presence of axion-like behavior [11][13] - The discovery of axion quasi-particles opens new avenues for applications in condensed matter physics and could lead to programmable quantum devices for detecting dark matter [13]
可视化模型为地震预警开辟新路径
Xin Hua She· 2025-06-18 07:38
在分析了26种不同的模拟地震情境后,研究人员发现破裂速度与断裂能(即"撕裂"或"裂开"材料所需的 能量)之间的关系符合线弹性断裂力学的预测。他们的计算机模拟不仅成功再现了实验中的缓慢和快速 地震过程,还在破裂速度、应力下降幅度和光透过率等多个维度与预测结果高度吻合。地震周期中接触 面积的变化会影响诸如电导率、渗透率和地震波透射率等多种可测物理属性。因此,持续监测这些间接 指标,有望揭示断层的行为变化。 这项研究结果揭示了一个长期隐藏的联系:标准地震模型中使用的经验性"状态变量",其实正对应着断 层面之间的实际接触面积。这是自20世纪70年代以来,地震科学领域第一次对这一关键数学变量给出物 理解释。 这一发现不仅具有理论意义,更可能为现实世界的地震预测带来突破。未来,监测断层真实接触状态的 物理属性可能成为短期预警系统的关键。(完) 美国南加州大学的研究人员介绍说,当两个粗糙表面相互滑动时,实际上只在极小的、孤立的接触点接 触,总接触面积只占整个表面的一小部分。而这一用肉眼无法看到、但可通过光学方法测量的实际接触 面积,正是控制地震行为的关键状态变量。 借助透明的丙烯酸材料,研究人员在实验室"真正看到"了地震破 ...
突破125年世纪难题!北大校友联手科大少年班才子破解希尔伯特第六问题
量子位· 2025-06-14 08:33
闻乐 发自 凹非寺 量子位 | 公众号 QbitAI 1900年,数学大师希尔伯特提出23个数学难题,其中第六个问题——"物理学的公理化",被称为数学物理的终极挑战。 125年后,北大校友邓煜、中科大少年班马骁与陶哲轩高徒扎赫尔・哈尼终于在这一问题上取得重大突破。 在20世纪,关于第六问题,希尔伯特追问: 能否像欧几里得几何一样,为物理学构建严格的数学基础? 因涉及从微观粒子动力学到宏观连续介质的多尺度关联,这个问题证明起来非常困难。 在微观层面,气体由无数粒子组成,单个粒子运动服从牛顿力学 (时间可逆) 。 在宏观层面,气体的统计行为由玻尔兹曼方程描述 (时间不可逆,趋向熵增) 。 该问题的核心目标是从弹性碰撞的硬球粒子系统出发,严格推导出流体力学的基本偏微分方程,完成希尔伯特第六问题中从原子论到连续介质 运动定律的推导程序。 解决该问题要分两步走,先通过 "动力学极限" 从牛顿定律推导出玻尔兹曼方程,再通过 "流体动力学极限" 从玻尔兹曼方程推导出流体方程。 如何从可逆的微观规律,演化出不可逆的宏观行为? 125年来,无数数学家在此领域折戟沉沙。 爱因斯坦的广义相对论、量子力学的数学框架虽部分实现了公理化 ...