编辑丨王多鱼 排版丨水成文 日前，西湖大学宣布， 纽约州立大学石溪分校 帝国创新教授、IEEE Fellow 凌海滨 全职加入西湖大学工学 院，受聘讲席教授，牵头创立 西湖大学 智能计算与应用实验室 ，该实验室致力于开展人工智能以及跨学 科的研究与应用，研究方向包括计算机视觉、多模态人工智能、增强现实、AI for Science、量子信息等方 向。 凌海滨 ， 1974 年生，贵州安顺人，国际电气和电子工程师协会会士 （IEEE Fellow） 。于 1997 年获北 京大学学士学位，2000 年获北京大学硕士学位，2006 年获美国马里兰大学帕克分校博士学位。其职业经 历包括微软亚洲研究院助理研究员 （2000–2001） 、加州大学洛杉矶分校博士后研究员 （2006– 2007） 、西门子研究院科学家 （2007–2008） 。2008 年起任教于天普大学，先后担任助理教授 （2008–2014） 和副教授 （2014–2019） 。2019-2025 年任纽约州立大学石溪分校计算机科学系 Empire Innovation 教授，2025 年加入西湖大学任人工智能讲席教授。 在北京大学期间， 凌海滨 ...

编辑丨王多鱼 排版丨水成文 肿瘤免疫微环境 （ Tumor Immune Microenvironment， TIME） 对癌症的发展进程和免疫治疗响应有着 至关重要的影响。 多重免疫荧光 （mIF） 是一种强大的成像方式，可用于解析肿瘤免疫微环境 （TIME） ，但其应用受到高昂成本和较低通量的限制。 近日， 微软研究院 潘海峰 、微软研究院/华盛顿大学 王晟 等人在国际大奖学术期刊 Cell 上发表了题为： Multimodal AI generates virtual population for tumor microenvironment modeling 的研究论文。 该研究提出了 GigaTIME ，这是一个通过连接细胞形态和状态来进行大规模群体肿瘤免疫微环境建模的多 模态人工智能框架 （该工具已在 Hugging Face、GitHub 和 Microsoft Foundry 免费开源） 。 GigaTIME 学习了一种跨模态翻译器，通过在 4000 万个细胞的配对 H&E 和 mIF 数据 （涵盖 21 种蛋白 质） 上进行训练，成功实现了 从常规 H&E 病理切片到多重免疫荧光 （ mIF ...

Core Insights - Artificial Intelligence (AI) is revolutionizing the trading landscape, bringing unprecedented efficiency, accuracy, and speed to financial markets. By 2025, AI is expected to handle nearly 89% of global trading volume, impacting everything from high-frequency stock trading to decentralized cryptocurrency ecosystems [1][10]. Group 1: Evolution of AI in Trading - The adoption of AI in trading is driven by the need for automation, reducing human errors, and executing trades at record speeds [10]. - Current markets generate and process over 2.5 million terabytes of data daily, including news, social media, satellite images, and blockchain transactions, creating a data explosion that AI can help manage [13]. Group 2: Core Technologies Driving AI Trading - Key technologies include machine learning, deep learning, natural language processing, and quantum computing, which enhance trading strategies and decision-making [13][14]. - AI systems can achieve nanosecond-level trading responses, significantly faster than human reaction times, improving overall market efficiency [13]. Group 3: AI Trading Platforms and Strategies - AI trading strategies encompass quantitative trading, algorithmic trading, sentiment analysis, and reinforcement learning, all aimed at maximizing returns and managing market risks [14][15]. - The regulatory landscape is evolving, with institutions like the SEC approving new AI-driven order types, legitimizing autonomous trading systems [13].

Core Insights - The research team from MIT has developed a multimodal robotic platform called CRESt, which significantly enhances the speed and quality of catalyst development by integrating multimodal models with high-throughput automated experiments [1][3][14] Group 1: Platform and Methodology - CRESt combines knowledge-assisted Bayesian optimization (KABO) with automated experiments to collect various forms of data within a unified active learning framework [3] - The platform utilizes precise control of chemical components, high-throughput scanning electron microscopy for microstructural imaging, and large language models to embed literature knowledge into the search space [6] - An innovative algorithm, Bayesian Optimization with Improved Constraints (BOPIC), dynamically adjusts the balance between exploration and exploitation, eliminating the need for manual parameter tuning [6] Group 2: Experimental Achievements - Within three months, CRESt completed over 900 catalyst chemical compositions and conducted more than 3,500 electrochemical tests, discovering formulations that significantly outperform traditional palladium-based catalysts [6][12] - The new eight-component high-entropy alloy catalyst demonstrated a 9.3-fold increase in unit cost power density compared to pure palladium benchmarks and achieved the highest performance in direct formate fuel cells with only a quarter of the previous noble metal loading [12] Group 3: Addressing Experimental Challenges - The research team tackled the common issue of reproducibility in experimental science, initially facing significant data noise due to inconsistencies in synthesis and testing [8] - Visual-language models (VLMs) were employed to diagnose sources of irreproducibility and suggest corrective measures, such as identifying misalignment in pipette tips and carbonized surfaces on sample holders [8][9] - The team improved stability and consistency by switching to stainless steel fixtures based on feedback from the VLM diagnostics [10] Group 4: Theoretical Insights - The study combined in situ X-ray absorption spectroscopy (XAS) with density functional theory (DFT) calculations to understand the mechanisms behind performance improvements [12] - Results indicated that palladium and platinum maintained metallic states under reaction conditions, which is crucial for catalytic activity, while dopants like Nb, Cr, and Ce introduced structural perturbations without significant lattice distortion [12] - DFT calculations revealed that the energy barrier for the rate-determining step in the indirect oxidation pathway was significantly lower in the high-entropy alloy compared to pure palladium, enhancing resistance to carbon monoxide poisoning [12]

Investment Rating - The report maintains a rating of "Accumulate" for the company [2][5] Core Viewpoints - The company achieved a significant revenue increase of 34.37% in 2024, reaching 737 million yuan, primarily driven by rapid growth in its security gateway and security service revenues, as well as breakthroughs in new security AI products [4][5] - Despite the revenue growth, the company reported a net loss of 119 million yuan in 2024 due to substantial increases in operating expenses, including a 122.45% rise in sales expenses and a 55.78% rise in management expenses [4][5] - The first quarter of 2025 saw a remarkable revenue growth of 444.91%, amounting to 308 million yuan, indicating strong performance in new business deliveries [4][5] Financial Projections - The company’s revenue projections for 2025, 2026, and 2027 are 814 million yuan, 922 million yuan, and 1,055 million yuan respectively, reflecting a growth rate of 10.5%, 13.2%, and 14.4% [2][4] - The forecasted net profits for 2025, 2026, and 2027 are 4 million yuan, 27 million yuan, and 61 million yuan respectively, showing a significant recovery from the previous losses [2][4] - The report highlights a projected increase in earnings per share (EPS) from -1.54 yuan in 2024 to 0.05 yuan in 2025, and further to 0.79 yuan in 2027 [2][4] Market Position and Strategy - The company is transitioning from being an innovator in network security visualization technology to becoming a builder of a secure computing ecosystem in the AI era [4][5] - Strategic partnerships have been established with various organizations to enhance technological development and market expansion, including collaborations with Inspur Cloud and Jiangyuan Technology [4][5] - The company has received recognition for its innovative products, including awards for its next-generation AI firewall and its inclusion in the digital security capability landscape by the China Academy of Information and Communications Technology [4][5]

Core Viewpoint - The Zhejiang Provincial Health Commission and ten other departments have issued an action plan to accelerate the high-quality development of "Artificial Intelligence + Healthcare" from 2025 to 2027, focusing on data resource aggregation and the establishment of a robust health data management system [1] Group 1: Data Management and Standards - The plan emphasizes the need to develop and improve data security management and utilization systems in the healthcare sector [1] - It aims to establish a health data standard system to enhance the quality and capacity of health data [1] - The initiative includes the construction of a trustworthy data space for the healthcare industry and the development of data governance tools and intelligent engines [1] Group 2: AI Development and Research - The action plan outlines the creation of a provincial medical bioinformatics database and the sharing of high-quality industry datasets and corpora [1] - It aims to build a multi-modal medical industry large model and establish a fully autonomous AI research and development framework [1] - The plan encourages the development of specialized models and medical intelligent agents covering various areas such as medical services, health management, public health, and drug/device research [1] Group 3: Interdisciplinary Collaboration and Innovation - The initiative focuses on deploying major technological projects in fields like AI data and applications, brain-computer interfaces, and new drug development [1] - It promotes interdisciplinary collaboration among research institutions to generate globally influential research outcomes in medical AI [1]

Core Viewpoint - The article discusses the transformative potential of artificial intelligence (AI) in the biomedical field, emphasizing advancements in large language models (LLMs) and multimodal AI that can enhance diagnostics, patient interactions, and medical predictions [2][6][11]. Group 1: Technological Innovations - Recent advancements in AI, particularly in LLMs and multimodal AI, are set to revolutionize the medical field by improving diagnostics and patient interactions [6]. - Key architectural innovations such as Transformer architecture, generative adversarial networks, and diffusion models have contributed to the development of complex generative AI systems [2][4]. Group 2: Medical Practice Transformation - AI-enabled medical practices are shifting clinical care from sporadic interactions to continuous monitoring and regular follow-ups, allowing for proactive healthcare in familiar environments [8]. - New medical knowledge can be more easily integrated into care models, and AI technologies are facilitating the development of new drugs [8]. Group 3: Multiscale Medical Predictions - AI algorithms can predict future medical events based on various dynamic inputs, applicable at multiple levels from molecular to population [10]. - The future of medicine will involve tools capable of processing vast amounts of information, significantly improving diagnostic accuracy and patient outcomes [11]. Group 4: Challenges and Implementation - Despite the promising advancements, the widespread clinical adoption of AI tools faces significant challenges, including bias, privacy concerns, regulatory hurdles, and integration with existing healthcare systems [6][11]. - Most AI tools are still in development, with few demonstrating clear benefits across all users or situations, which remains a major barrier to broader usage by healthcare professionals [11]. Group 5: Roadmap for AI Implementation - The roadmap for implementing medical AI involves transitioning from basic scientific research to concept validation models, leading to larger models and early clinical applications that pave the way for final clinical deployment and optimization [14].

Core Viewpoint - Sudden Cardiac Death (SCD) is a major global health issue, particularly in patients with Hypertrophic Cardiomyopathy (HCM), where current clinical guidelines show low performance and inconsistent accuracy in risk assessment [1][2]. Group 1 - SCD has an annual incidence of 50-100 cases per 100,000 people in North America and Europe, with ventricular arrhythmias being the primary mechanism [1]. - Implantable Cardioverter Defibrillators (ICDs) can effectively terminate arrhythmias and reduce mortality in high-risk patients when implanted preventively [1]. - The current risk stratification paradigm, based on Left Ventricular Ejection Fraction (LVEF) below 30%-35%, is primarily applicable to ischemic and dilated cardiomyopathy patients but fails to provide comprehensive risk assessment [2]. Group 2 - A recent study published by researchers from Johns Hopkins University introduced a multimodal AI model named MAARS (Multimodal Artificial intelligence for Ventricular Arrhythmia Risk Stratification) to predict arrhythmic death in HCM patients [3][4]. - MAARS utilizes a Transformer-based neural network that learns from electronic health records, echocardiograms, radiology reports, and contrast-enhanced cardiac MRI, which is a unique aspect of the model [8]. Group 3 - MAARS demonstrated an Area Under the Curve (AUC) of 0.89 in internal cohorts and 0.81 in external cohorts, significantly outperforming current clinical guidelines which have AUCs ranging from 0.27-0.35 (internal) and 0.22-0.30 (external) [10]. - Unlike clinical guidelines, MAARS shows fairness across different population subgroups, enhancing the transparency of AI predictions and identifying risk factors for further investigation [10]. - Overall, MAARS is a powerful and reliable clinical decision support tool for risk stratification of SCD in HCM patients, with potential to significantly improve clinical decision-making and patient care through integration with automated data extraction systems or as a concept validation for personalized patient care [10].

Core Insights - Alibaba Cloud has achieved the top position in China's AI infrastructure market with a 23% market share, surpassing the combined share of the second and third players [1] - In the generative AI infrastructure sector, Alibaba Cloud has secured both model training and inference market leadership [1] Group 1: AI Infrastructure Market Position - According to IDC's latest report, Alibaba Cloud leads the AI IaaS market in China with a 23% share [1] - The company's dominance in the generative AI infrastructure is highlighted by its dual championship in model training and inference [1] Group 2: Qwen VLo Model Features - The Qwen VLo model, launched on June 27, features a multi-modal understanding and generation capability, allowing users to experience it through Qwen Chat [2] - Qwen VLo employs a progressive generation method, constructing images step-by-step while optimizing predictions, ensuring harmony in structure, color, and semantics [3] - The model supports dynamic resolution training, enabling image generation at any resolution and aspect ratio, breaking previous limitations [3] Group 3: User Experience and Flexibility - Qwen VLo enhances semantic consistency by accurately recognizing and preserving key features during image modifications, such as color changes [4] - Users can issue creative commands in natural language, allowing for artistic style transfers and complex scene modifications [4] - The model supports multiple languages, enabling global users to generate images based on simple descriptions in their preferred language [5] Group 4: Applications in Various Industries - In advertising design, Qwen VLo can quickly generate multiple design drafts based on user input, significantly reducing design cycles [5] - In education, the model can visualize abstract concepts, aiding teachers in explaining geographical features and literary themes [6] - The model also assists in game development by generating art resources and allows players to create personalized game elements [6]

Group 1 - The sixth Multinational Corporation Leaders Summit opened in Qingdao, emphasizing China's commitment to high-level opening-up and creating a world-class business environment, welcoming global companies to invest in China [1] - The Ministry of Commerce announced that it will expedite the review of export license applications related to rare earths, indicating a regulatory focus on this critical industry [1] - A video conference was held by the Ministry of Industry and Information Technology and other departments to strengthen safety management in the new energy vehicle sector, stressing the importance of maintaining product quality and avoiding short-term cost-cutting measures [1] Group 2 - The solar industry is expected to see a significant production cut in the third quarter, with operating rates projected to decrease by 10%-15% compared to the previous quarter [2] - Shanghai Guozhi Technology Co., Ltd. was established to create a competitive asset management service platform, involving major financial enterprises, which will support Shanghai's goal of becoming a global asset management center [2] - Pop Mart's Labubu products have seen a significant drop in second-hand prices following a large restock, indicating market volatility and the impact of supply on pricing [2] Group 3 - Zhongyan Chemical announced plans for a significant asset restructuring involving a reduction in capital by shareholders of Zhongyan Alkali Industry [3] - Shengnuo Bio is expected to report a substantial increase in net profit for the first half of the year, with projections of a 254%-332% year-on-year growth [3] - Zongsheng Pharmaceutical's subsidiary has completed the first participant enrollment for a Phase III clinical trial of its innovative drug RAY1225 for obesity/overweight [3] Group 4 - Lianchuang Optoelectronics indicated strong demand for drone countermeasure equipment in the Middle East, actively seeking local customers and partners [3] - CloudWalk Technology stated that its multi-modal AI technology for live detection and dynamic verification can be adapted for stablecoin wallet scenarios [3]