证券研究报告 | 2025年10月11日 多因子选股周报 超额全线回暖，四大指增组合本周均跑赢基准 核心观点 金融工程周报 国信金工指数增强组合表现跟踪 因子表现监控 以沪深 300 指数为选股空间。最近一周，预期 EPTTM、一个月波动、BP 等因子表现较好，而单季营收同比增速、三个月机构覆盖、3 个月盈利上下 调等因子表现较差。 以中证 500 指数为选股空间。最近一周，SPTTM、预期 BP、单季 EP 等因 子表现较好，而一年动量、预期 PEG、标准化预期外收入等因子表现较差。 以中证 1000 指数为选股空间。最近一周，EPTTM、SPTTM、预期 EPTTM 等因子表现较好，而预期净利润环比、一年动量、单季营收同比增速等因子 表现较差。 以中证 A500 指数为选股空间。最近一周，单季 SP、SPTTM、一个月波动 等因子表现较好，而单季营收同比增速、一年动量、三个月机构覆盖等因子 表现较差。 以公募重仓指数为选股空间。最近一周，预期 EPTTM、单季 EP、一个月波 动等因子表现较好，而一年动量、单季营收同比增速、预期净利润环比等因 子表现较差。 公募基金指数增强产品表现跟踪 目前，公募基金沪深 ...

GPT-5首次通过「哥德尔测试」，连破三大组合优化猜想！甚至，它能自主推翻原有猜想，给出全新有效解法，当场惊呆OpenAI研究科学家。 AI迎来历史性一刻！ GPT-5成功破解三大猜想，通过了「哥德尔测试」。 论文地址：https://arxiv.org/pdf/2509.18383 OpenAI科学家Sebastien Bubeck惊叹地表示，这类开放性问题，顶尖博士生往往耗费数日才能解决。 不同以往，这项由海法大学和思科主导的研究，首次让AI直面「开放性数学猜想」的挑战。 论文中，团队设计了五项「组合优化」领域的测试任务，每项任务提供1-2篇文献作为了解。 在三个相对简单的问题上，GPT-5给出了近乎完美的解法，证明了其强大的逻辑推理水平。 (1) = (1 = $\varepsilon$) = max f(T), TET (1) = (1 = $\varepsilon$) = (1 = $\varepsilon$) = max f(T), TET (1) = (1 = $\varepsilon$) = (1 = $\varepsilon$) = max f(T), TET (1) = (1 = $\va ...

Core Insights - The article discusses the groundbreaking research conducted by Professor Yan Junchi's team at Shanghai Jiao Tong University, focusing on advancements in AI, robotics, and autonomous driving [2][32]. - The team's recent publications in top conferences like CVPR, ICLR, and NeurIPS highlight key trends in AI research, emphasizing the integration of theory and practice, the transformative impact of AI on traditional scientific computing, and the development of more robust, efficient, and autonomous intelligent systems [32]. Group 1: Recent Research Highlights - The paper "Grounding and Enhancing Grid-based Models for Neural Fields" introduces a systematic theoretical framework for grid-based neural field models, leading to the development of the MulFAGrid model, which achieves superior performance in various tasks [4][5]. - The "CR2PQ" method addresses the challenge of cross-view pixel correspondence in dense visual representation learning, demonstrating significant performance improvements over previous methods [6][7]. - The "BTBS-LNS" method effectively tackles the limitations of policy learning in large neighborhood search for mixed-integer programming (MIP), showing competitive performance against commercial solvers like Gurobi [8][10][11]. Group 2: Performance Metrics - The MulFAGrid model achieved a PSNR of 56.19 in 2D image fitting tasks and an IoU of 0.9995 in 3D signed distance field reconstruction tasks, outperforming previous grid-based models [5]. - The CR2PQ method demonstrated a 10.4% mAP^bb and 7.9% mAP^mk improvement over state-of-the-art methods after only 40 pre-training epochs [7]. - The BTBS-LNS method outperformed Gurobi by providing a 10% better primal gap in benchmark tests within a 300-second cutoff time [11]. Group 3: Future Trends in AI Research - The research indicates a shift towards a deeper integration of theoretical foundations with practical applications in AI, suggesting a future where AI technologies are more robust and capable of real-world applications [32]. - The advancements in AI research are expected to lead to smarter robots, more powerful design tools, and more efficient business solutions in the near future [32].

Core Insights - The article discusses the rapid development of scientific research and industrial applications driven by artificial intelligence (AI) and optimization, while highlighting the significant energy consumption challenges these technologies pose for sustainable digital computing [1][2]. Group 1: Analog Optical Computer (AOC) - The Microsoft Cambridge Research team proposed the Analog Optical Computer (AOC), which can efficiently perform AI inference and optimization tasks without frequent digital conversions, offering significant scalability and energy efficiency advantages [3][5]. - AOC combines analog electronic technology with 3D optical technology, enabling a dual-domain capability that enhances noise resistance and supports recursive reasoning in computationally intensive neural models [5][7]. - The AOC architecture is built on scalable consumer-grade technology, providing a promising path for faster and more sustainable computing [7][18]. Group 2: Applications and Performance - AOC is primarily aimed at two types of tasks: machine learning inference and combinatorial optimization, with the research team demonstrating its capabilities through four typical case studies [8]. - In machine learning tasks, AOC successfully executed image classification and nonlinear regression, achieving higher accuracy compared to traditional linear classifiers [9]. - For combinatorial optimization, AOC demonstrated its effectiveness in medical image reconstruction and financial transaction settlement, achieving accurate results without any digital post-processing [10][11]. Group 3: Scalability and Efficiency - AOC is expected to support models with parameter scales ranging from 100 million to 2 billion, requiring between 50 to 1000 optical modules for operation [16][17]. - The estimated power consumption for processing a matrix with 100 million weights using 25 AOC modules is 800 W, achieving a computational speed of 400 Peta-OPS, with energy efficiency of 500 TOPS per watt [17]. - AOC's architecture shows potential for achieving approximately 100 times energy efficiency improvement in practical machine learning and optimization tasks [18][19].

Core Viewpoint - The small-cap style continues to show strength in the market, with the CSI 2000 Enhanced ETF (159552) and the 1000 ETF Enhanced (159680) both reaching new highs since their listing, driven by macroeconomic trends and industry upgrades [1][2][5]. Group 1: Small-Cap Style Performance - The CSI 2000 Enhanced ETF (159552) achieved a net value growth rate of 29.18% in the first half of the year, ranking first among broad-based ETFs, with an excess return of nearly 14% [1]. - The small-cap index turnover rate was 2.1% as of June 27, indicating a relatively high trading congestion level, while the small-cap to large-cap index turnover ratio was approximately 4.1 times, close to historical averages [5]. - The current price-to-earnings (P/E) ratio of the small-cap index to the large-cap index is 2.2 times, positioned at the 72.5% percentile since 2015, suggesting a favorable valuation environment for small-cap stocks [5]. Group 2: Macroeconomic and Industry Trends - The macroeconomic direction and industry upgrade trends are key signals for the rotation between small and large-cap stocks, with small-cap stocks showing relative advantages during periods of technological innovation and policy encouragement [2][4]. - The ongoing favorable environment for small-cap stocks is supported by the thriving sectors of AI and semiconductors, as well as continued policy support for the development of new productive forces [5]. Group 3: Enhanced ETF Performance - The CSI 2000 Enhanced ETF (159552) has consistently delivered excess returns since its establishment on June 29, 2024, with each quarter showing excess returns exceeding 6% in the first two quarters of this year [6]. - The 1000 ETF Enhanced (159680) has also demonstrated significant enhancement effects, achieving a cumulative excess return of 33.10% since its inception on November 18, 2022, with an annualized excess return of 11.88% [9][11]. - Both enhanced ETFs have shown strong adaptability to different market conditions, capturing excess returns during both downward trends and upward surges [8][11].

金磊 发自 凹非寺 量子位 | 公众号 QbitAI 天下苦大模型 矩阵乘法 久矣。 毕竟不论是训练还是推理过程，矩阵乘法作为最主要的计算操作之一，往往都需要消耗大量的算力。 那么就没有一种更"快、好、省"的方法来搞这事儿吗？ 有的， 香港中文大学 最新一篇仅 10页 的论文，便提出了一种新算法： 论文作者之一的Dmitry Rybin表示： 这项研究对数据分析、芯片设计、无线通信和LLM训练都有着深远的影响！ 能源可节省：5%-10% 时间可节省：5% 这么算矩阵乘法，更快！ 矩阵乘法是计算机科学和数值线性代数中的核心问题之一。 自从Strassen和Winograd的开创性工作以来，研究者们一直在探索如何减少矩阵乘法所需的计算量。 尽管这类运算在统计、数据分析、深度学习和无线通信等领域有着广泛应用，例如协方差矩阵的计算和线性回归中的关键步骤，但对于具有 特殊结构的矩阵乘法（如计算矩阵与其转置的乘积XX t ）的研究相对较少。 从理论角度看，计算XX t 与一般矩阵乘法具有相同的渐近复杂度，因此只能通过常数因子优化来提升速度。 因此，这篇论文《XX t Can Be Faster》提出了一种名为RXTX的新 ...