Quantitative Models and Construction Methods Model Name: ML-Master - **Model Construction Idea**: The ML-Master model is designed to simulate human expert cognitive strategies, addressing the three major bottlenecks in existing AI4AI systems: low exploration efficiency, limited reasoning ability, and module fragmentation[12] - **Model Construction Process**: - **Balanced Multi-Trajectory Exploration Module**: Utilizes a parallelized Monte Carlo tree search to model the AI development process as a dynamic decision tree, with each node representing a potential solution state. This module dynamically allocates computing resources based on the potential value of 75 Kaggle task branches, avoiding local optima and improving medium difficulty task medal rates to 20.2%, 2.2 times the baseline method[13] - **Controllable Reasoning Module**: Overcomes the static decision limitations of large language models by filtering key code fragments, performance metrics, and cross-node insights from historical explorations through an adaptive memory mechanism. This ensures the reasoning process is based on verifiable execution feedback rather than probabilistic guesses, improving high difficulty task performance by 30%, significantly surpassing Microsoft's system's 18.7%[13] - **Adaptive Memory Mechanism**: Integrates the exploration and reasoning modules, creating a closed-loop evolution system. The results of code execution collected during the exploration phase are embedded into the reasoning model's "think" phase after intelligent filtering, and the optimized solutions from the reasoning output guide subsequent exploration paths. This dual empowerment allows ML-Master to reach the Grandmaster level among the top 259 global Kaggle participants after 900 machine hours of training, with solution quality improving by 120% over multiple iterations[15] - **Model Evaluation**: The ML-Master model demonstrates significant advantages in exploration efficiency, reasoning ability, and module integration, making it a leading system in the AI4AI field[12][13][15] Model Backtesting Results - **ML-Master**: - **Average Medal Rate**: 29.3%[12] - **Effective Submission Rate**: 93.3%[19] - **Task Performance**: 44.9% of tasks outperform more than half of human participants, with 17.3% of tasks winning gold medals[19] Quantitative Factors and Construction Methods Factor Name: OpenEvolve - **Factor Construction Idea**: OpenEvolve is designed to autonomously evolve code, achieving significant performance improvements in GPU kernel optimization tasks[22] - **Factor Construction Process**: - **Algorithm Layer**: Through 25 generations of evolutionary iterations, OpenEvolve autonomously discovered three key optimization strategies. For example, the SIMD optimization for Apple Silicon demonstrated the system's precise grasp of hardware characteristics, perfectly matching the hardware's SIMD width when processing 128-dimensional attention heads[23] - **Technical Implementation**: Utilizes a multi-model collaborative evolutionary architecture. The main model, Gemini-2.5-Flash, is responsible for rapid exploration, while the auxiliary model, Gemini-2.5-Pro, performs deep optimization. The system divides the Metal kernel function source code into evolvable blocks, retaining the integration code with the MLX framework unchanged, and evolves five subpopulations in parallel using the island model, with each generation having a population size of 25 individuals[24] - **Performance Evaluation**: The evaluation phase adopts a high-robustness design, including Metal command buffer protection, memory access violation handling, and exponential backoff retry mechanisms, ensuring the system can boldly attempt aggressive optimizations without worrying about crashes[25] - **Factor Evaluation**: OpenEvolve redefines the boundary of human-machine collaboration, demonstrating the potential for AI to autonomously explore optimization paths that require deep professional knowledge[22][23][24] Factor Backtesting Results - **OpenEvolve**: - **Average Performance Improvement**: 12.5% in decoding speed, 14.4% in pre-filling speed, and 10.4% in overall throughput[25] - **Peak Performance Improvement**: 106% in decoding speed for repetitive pattern generation tasks[25] - **Accuracy and Error Rate**: Maintains 100% numerical accuracy and zero GPU errors[25]

证券研究报告 | 行业周报 gszqdatemark 2025 07 07 年 月 日 传媒 百度文心大模型 4.5 系列开源，字节发布图像生成新模型 Xverse 行情概览：本周（6.30-7.4）中信一级传媒板块上涨 2.39%。本周传媒板 块在游戏板块带动下继续上涨，临近中报期重视中报预期较好公司的投资机 会。2025 年下半年传媒继续看好游戏等基本面驱动板块，同时弹性方向看好 AI 应用及 IP 变现。AI 应用聚焦新应用的映射投资及部分较成熟应用的数据跟 踪，重点关注 AI 陪伴、AI 教育及 AI 玩具方向。IP 变现聚焦有 IP 优势及全产 业链潜力的公司，潮流玩具、影视内容等方向有机会。 板块观点与关注标的：1）游戏：重点关注 ST 华通、吉比特、恺英网络、巨 人网络、神州泰岳、心动公司等，关注完美世界、冰川网络、华立科技等；2） AI：豆神教育、盛天网络、上海电影、荣信文化、盛天网络、中文在线、易点 天下、视觉中国、盛通股份、焦点科技、世纪天鸿、佳发教育等；3）资源整 合预期：中视传媒、国新文化、广西广电、华智数媒、吉视传媒、游族网络 等；4）国企：慈文传媒、皖新传媒、中文传媒、南方传媒、凯 ...

Group 1 - The report highlights a positive investment strategy for the internet industry, recommending a focus on independent growth logic in AIGC and the music industry during a period of index fluctuations [1][3] - In June, the Hang Seng Technology Index recorded a 2.5% increase, while the Nasdaq Index rose by 6.8% [11] - The report notes that major internet companies saw stock price increases, with Meituan, Kingdee International, and Kingsoft being the top performers in Hong Kong stocks, achieving monthly gains of 38.2%, 25.5%, and 23% respectively [14] Group 2 - The report indicates that in the gaming sector, the number of game licenses approved in June reached a two-year high, with 158 licenses issued [40] - In the fintech sector, the reserve funds of payment institutions increased by 8.4% year-on-year in May, reaching 24,573.54 billion yuan [42] - The e-commerce sector is experiencing intense competition, with platforms continuing to offer benefits to merchants and increasing investments in instant retail to seek new growth [3][47] Group 3 - Tencent Music's acquisition of Ximalaya was finalized, with a total transaction value of approximately 12.6 billion USD [45] - The report mentions that Alibaba's international station saw a 42% year-on-year increase in orders from June to date, with GMV maintaining a nearly 30% strong growth [53] - The report emphasizes that the AI sector is benefiting from major companies' business scenarios, such as cloud computing and advertising, while short-term AI agents still require refinement [3]

作者 | 褚杏娟 6 月 30 日，百度正式开源文心大模型 4.5 系列模型，涵盖 47B、3B 激活参数的混合专家（MoE） 模型，与 0.3B 参数的稠密型模型等 10 款模型，并实现预训练权重和推理代码的完全开源。 目前，文心大模型 4.5 开源系列已可在飞桨星河社区、HuggingFace 等平台下载部署使用，系列权 重按照 Apache 2.0 协议开源，同时开源模型 API 服务也可在百度智能云千帆大模型平台使用。值得 关注的是，此次文心大模型 4.5 系列开源后，百度实现了框架层与模型层的"双层开源"。 相关链接： https://huggingface.co/models?other=ERNIE4.5 https://aistudio.baidu.com/modelsoverview | (0) STEAR FILE CARDER CONSULT CARDER STATUS CONSTITUTION | A SETHA FILIPE A CONTRACT AND THE WARD I I | | --- | --- | | Image-Text-to-Text · . . : 424B · U ...

市场研究部 2025 年 2 月 24 日 计算机行业周度观察 本周计算机行业指数表现 本周（2.17-2.23）计算机（申万）板块上涨 3.17%，沪深 300 指数 上涨 1.00%，计算机板块跑赢沪深 300 2.17 个百分点。和申万其他行 业对比，计算机行业涨幅排名位列第 5 位。 本周涨幅前 3 名分别为东华软件(+48.88%)、宏景科技(+36.31%)、索 辰科技(+31.62%)，跌幅前 3 名分别为上海钢联(-10.03%)、永信至诚 (-10.87%)、新炬网络(-16.47%)。 行业动态 人工智能 华为擎云为代表的新势力在迅速落地自己研发的数字教育解决方案。 华为擎云与成都市七中育才学校（东湖校区）携手，成立华为擎云 智慧教育自主创新实训室，并成功落地擎云智慧教育-计算机教室解 决方案。擎云智慧教育-计算机教室解决方案的又一大亮点是分布式 桌面云也是华为擎云计算机教室解决方案的重要特色之一。通过网 络同传、运维分析系统，运维人员可以对设备远程集中管理维护， 数据来源：Wind，国新证券整理 邮箱：wangwenyj@crsec.com.cn 分析师：王闻 登记编码：S1490519 ...