来自伦敦大学学院（UCL）计算机科学系的研究团队提出了一种用于白细胞形态分析的联邦学习框架，使各机构能够在不交换训练数据的情况下进行协同 训练。利用来自多个临床站点的血液涂片，该联邦模型在保证完全数据隐私的同时，学习到稳健且域不变的特征表示。与集中式训练相比，联邦训练在跨 站点性能和对未知机构的泛化能力上表现出色。 血液形态学检查是临床诊断血液疾病的重要环节，通过观察外周血涂片（PBS）或骨髓穿刺（BMA）中的细胞形态，医生可以判断白血病、贫血、感染 及遗传性血液疾病的类型。然而，这一过程不仅劳动强度大，而且高度依赖经验丰富的专业人员。尤其在低收入和中等收入国家（LMICs），技能专家稀 缺，使得快速、可靠且可扩展的血液学诊断成为急需解决的问题。 近年来，人工智能和深度学习的发展为血液形态分析提供了新的解决方案。AI 模型能够自动识别不同类型的白细胞，并辅助医生进行快速诊断。研究表 明，深度学习在自动化血液学诊断中具备显著潜力，但现实应用中仍面临重要挑战——模型训练对数据的依赖性极强，而临床数据通常分布在不同医院， 且存在染色方法差异、成像设备差异以及少数罕见细胞类型的问题。这种数据异质性会导致模型在新机构或 ...

天眼查资料显示，中移（上海）信息通信科技有限公司，成立于2018年，位于上海市，是一家以从事科 技推广和应用服务业为主的企业。企业注册资本200000万人民币。通过天眼查大数据分析，中移（上 海）信息通信科技有限公司参与招投标项目2020次，财产线索方面有商标信息34条，专利信息953条， 此外企业还拥有行政许可6个。 中移智行网络科技有限公司，成立于2015年，位于上海市，是一家以从事软件和信息技术服务业为主的 企业。企业注册资本100000万人民币。通过天眼查大数据分析，中移智行网络科技有限公司参与招投标 项目705次，财产线索方面有商标信息106条，专利信息861条，此外企业还拥有行政许可8个。 中国移动通信集团有限公司，成立于1999年，位于北京市，是一家以从事电信、广播电视和卫星传输服 务为主的企业。企业注册资本30000000万人民币。通过天眼查大数据分析，中国移动通信集团有限公司 共对外投资了55家企业，参与招投标项目5000次，财产线索方面有商标信息2211条，专利信息5000条， 此外企业还拥有行政许可50个。 国家知识产权局信息显示，中移（上海）信息通信科技有限公司;中移智行网络科技有限公 ...

Group 1 - The 2025 Brand Strong Country Economic Forum was held in Beijing, where the group standard "Optimizing Consumption Environment and Reassuring Consumption Brand Evaluation Norm" was officially launched [1] - The standard was drafted by the National Business Newspaper Association in collaboration with authoritative institutions, establishing a comprehensive evaluation index system for industry standardization and reassuring consumption brand construction [1] - Aier Eye Hospital Group, as one of the drafting units, integrates its development experience into the standard to assist in optimizing the consumption environment and building a reassuring consumption ecosystem [1] Group 2 - Aier Eye Hospital has created a tiered chain model to address the uneven distribution of medical resources, promoting a three-dimensional eye care service network [2] - The company aims to synchronize medical technology, equipment, and pharmaceuticals with international standards, ensuring that innovative technologies benefit Chinese eye care patients sooner [2] - Aier Eye Hospital has achieved steady growth in outpatient volume, surgical volume, and discharge numbers, with a postoperative infection rate of 0.0156% for high-level surgeries, outperforming international averages [2] Group 3 - Aier Eye Hospital is advancing its digital transformation in eye care by launching a "Digital Eye Care" model, utilizing cutting-edge technologies like AI and federated learning [3] - The company aims to create an intelligent closed-loop management system for eye health services, enhancing accessibility and efficiency for grassroots patients [3] - Aier Eye Hospital plans to use the new standard as a guide to deepen technological innovation and upgrade service quality, contributing to the high-quality development of the Chinese eye care industry [3]

Core Insights - The founder and chairwoman of the company participated in the World Economic Forum's 56th Annual Meeting, engaging in high-level discussions on the intelligent transformation of the healthcare industry and the role of AI in addressing global health challenges [1][3][5]. Group 1: AI and Healthcare Resilience - The company emphasized the importance of AI and big data in enhancing public health system resilience and addressing new health challenges driven by climate change [1]. - AI healthcare enterprises can leverage technology to monitor climate-sensitive disease trends and innovate chronic disease management, thereby improving societal health resilience [1]. Group 2: AI in Health Security - The company highlighted the critical role of AI in infectious disease monitoring, drug development, and health management during discussions on reconstructing health security [3]. - The use of federated learning and privacy computing technologies is being implemented to enhance data security while improving model training efficiency across regions [3]. Group 3: China's AI Healthcare Development - The company discussed China's unique advantages in AI healthcare, including rich application scenarios, an improving data ecosystem, and strong policy support [5]. - The global value of China's experience lies not only in technological breakthroughs but also in systematic implementation capabilities and efficient localization models [5]. Group 4: Global Collaboration and Innovation - The company is actively building and expanding its collaboration network through bilateral meetings with key stakeholders, including government officials and industry leaders [7]. - A significant partnership was announced with Novartis for a cardiovascular disease prevention project, showcasing the company's commitment to AI-driven healthcare solutions [7]. Group 5: Contribution to Global AI Healthcare Ecosystem - The company's presence at the forum underscored the role of new-generation Chinese tech entrepreneurs in contributing to global wisdom and fostering cross-sector collaboration for a resilient future [8].

Core Viewpoint - The article discusses the vibrant developments in China's AI sector at the beginning of 2026, highlighting key figures in the field and their contributions to the evolution of large models and AI applications. Group 1: Event Highlights - The event featured prominent figures in AI, including Professor Tang Jie, Yang Zhilin, Lin Junyang, and Yao Shunyu, marking a significant gathering in Beijing [1]. - The presence of foundational figures like Zhang Bo and Yang Qiang indicates the event's importance in shaping the future of the large model industry [1]. Group 2: Observations on AI Development - The year 2025 was noted as a breakthrough year for open-source models in China, with a 10 to 20 times increase in coding activities [6]. - The discussion emphasized the differentiation of AI models, with a focus on enterprise applications and coding, inspired by developments in Silicon Valley [7][8]. Group 3: Model Differentiation - Yao Shunyu pointed out the clear division between To C (consumer) and To B (business) models, with a growing trend towards vertical integration and layered applications [9][12]. - The article highlights that while consumer applications may not require the highest intelligence, business applications benefit significantly from stronger models, leading to a willingness to pay for superior performance [10][11]. Group 4: Future Paradigms in AI - The conversation shifted to the next paradigm in AI, focusing on autonomous learning and self-improvement, with various interpretations of what this entails [23][24]. - Yao Shunyu mentioned that the bottleneck for autonomous learning is not methodology but rather the data and tasks involved, indicating a need for context and environment to enhance AI capabilities [23][25]. Group 5: Agent Strategy - The potential for agents to automate human tasks significantly was discussed, with expectations that by 2026, agents could handle workloads equivalent to one or two weeks of human effort [39][40]. - The article suggests that the development of agents is closely tied to advancements in model capabilities and the complexity of interaction environments [45][46].

Core Viewpoint - Federated Learning (FL) is not as secure as previously thought, as Gradient Inversion Attacks (GIA) can potentially compromise privacy by reconstructing private training data from shared gradient information [3][5]. Group 1: Background and Importance of the Study - Federated Learning allows clients to collaboratively train models without sharing raw data, but recent studies indicate that "not sharing data" does not equate to "absolute security" [5]. - Attackers can utilize GIA to reconstruct private data such as facial images and medical records, highlighting the need for a systematic classification and analysis of these attacks [5][6]. Group 2: Classification of GIA Methods - The research categorizes existing GIA methods into three main types: 1. Optimization-based attacks (OP-GIA) 2. Generation-based attacks (GEN-GIA) 3. Analysis-based attacks (ANA-GIA) [9]. Group 3: Theoretical Contributions - The study presents significant theoretical advancements, including: - Theorem 1: Establishes a linear relationship between the reconstruction error of OP-GIA and the square root of Batch Size and image resolution, indicating that larger batch sizes and higher resolutions make attacks more difficult [11]. - Proposition 1: Reveals that the similarity of gradients during model training affects the difficulty of data recovery, with more similar gradients making recovery harder [13]. Group 4: Experimental Findings - Extensive experiments were conducted on datasets like CIFAR-10/100, ImageNet, and CelebA, covering various attack types and model architectures [15]. - Key findings indicate that: - OP-GIA is practical but limited by batch size and resolution, with its threat significantly reduced in Practical FedAvg scenarios. - GEN-GIA can generate high-quality images but relies heavily on pre-trained generators and specific activation functions, making it less effective if those conditions are not met. - ANA-GIA can achieve precise data recovery but is easily detectable by clients, limiting its practical application [25]. Group 5: Defense Guidelines - The authors propose a three-phase defense pipeline to enhance security without complex encryption: 1. Network design phase 2. Training protocol phase 3. Client verification phase, where clients should validate model architecture and parameters to prevent malicious modifications [22]. Group 6: Summary and Practical Implications - This research serves as a comprehensive examination of existing GIA methods and provides practical guidelines for enhancing the security of federated learning systems, emphasizing that while privacy risks are real, they can be effectively managed through thoughtful design and protocols [24].

Core Insights - Haier Consumer Finance successfully concluded its 2025 "Feature Hero" initiative, aimed at enhancing data-driven value in financial services and expanding multi-dimensional data samples [1][6] - The initiative emphasizes the importance of data and features in risk control, with advanced models and algorithms striving to approach the risk identification "ceiling" determined by data [1] Group 1: Feature Hero Competition - The first prize of the "Feature Hero" competition was awarded to the Risk Management Center, which innovatively utilized large models to replace manual processing of voice data, aiding in credit risk control strategies [5] - The competition attracted 32 employees, resulting in the extraction of 2,023 high-quality features from vast data, significantly enhancing the risk control system [5] Group 2: Intelligent Risk Control System - By 2025, Haier Consumer Finance's intelligent risk control system had launched a total of 10,427 real-time features, a 70% increase year-on-year [6] - The company emphasizes the importance of continuous competitions like "Feature Hero" to foster an AI-driven culture and enhance data asset exploration [6] Group 3: AI Integration and Industry Trends - The integration of deep learning technologies such as large models, graph learning, and natural language processing is transforming credit risk control models, showcasing a trend of multi-technology application in the field [6] - Haier Consumer Finance's AI-driven risk control system significantly reduces fraud risk and improves credit approval efficiency, achieving a dual advantage of controllable risk and efficient service [6] Group 4: Future Developments - Future advancements in technologies like federated learning, reinforcement learning, and AGI are expected to further enhance risk control models in areas such as data privacy protection and dynamic strategy optimization [7] - The company plans to deepen its AI First strategy, continuously strengthening data governance and technical application capabilities for high-quality development in credit business [7]

Core Viewpoint - The establishment of the PolyU Artificial Intelligence Advanced Research Institute (PAAI) aims to address challenges in AI development, focusing on decentralized AI architectures and the transformation of research outcomes into practical applications [1][2]. Group 1: PAAI Establishment and Goals - PAAI is one of the five key research institutes at Hong Kong Polytechnic University, dedicated to interdisciplinary research and the conversion of cutting-edge research into products and technologies [1]. - The institute aims to leverage PolyU's research network in mainland China to facilitate the application of AI technologies for societal development [1]. Group 2: Government Support and AI Ecosystem - The Hong Kong government is advancing AI development through talent acquisition, data management, and application initiatives, with plans to establish the Hong Kong AI Research Institute (AIRDI) by 2026 [1]. - Over 1,000 experts in AI and robotics have been gathered in Hong Kong, highlighting the region's commitment to fostering AI growth [1]. Group 3: Addressing AI Challenges - PAAI aims to tackle issues such as data centralization and privacy concerns, promoting a shift from centralized to decentralized AI models [2]. - The concept of "Federated Learning" is introduced as a secure, decentralized approach to AI development, enabling data flow across sectors like healthcare and finance without compromising privacy [2]. Group 4: Technological Innovations and Collaborations - PAAI is focusing on collaborative generative AI, federated learning, and edge-based models, with applications in healthcare, education, finance, and robotics [2]. - The institute is collaborating with medical institutions on the "Cancer GenAI" project and exploring AI applications in infectious disease control and financial sectors [2][3]. Group 5: Co-GenAI Project and Funding - The "Co-GenAI" project aims to develop decentralized AI technologies, addressing the limitations of centralized AI in accessing private data in high-end fields like healthcare [3]. - PAAI has received dual funding from the Hong Kong government's RAISE+ and TRS programs to support the "Cancer GenAI" project and has launched a blockchain-based global research collaboration platform [3].

Core Viewpoint - The article highlights the evolution and significance of the GAIR (Global Artificial Intelligence and Robotics Conference) as a platform for Chinese AI scholars, particularly those associated with Microsoft, to connect and collaborate, marking a shift in China's position in the global AI landscape [5][9]. Group 1: Historical Context - In 1996, Wu Feng, a doctoral student at Harbin Institute of Technology, reached out to Zhang Yaqin, a prominent scientist, to advocate for China's inclusion in the MPEG committee, aiming to enhance the international recognition of local scholars [2][4]. - Zhang Yaqin, alongside Li Kaifu, co-founded the Microsoft Research Asia, which became a pivotal institution for AI development in China, fostering connections between academia and industry [5][6]. Group 2: GAIR Development - The first GAIR conference was held in Shenzhen, initiated by prominent figures like Zhu Xiaorui and Lin Jun, bringing together top overseas scientists to discuss AI and robotics [7][8]. - Over the years, GAIR has become a gathering point for over 40 Microsoft-affiliated scientists, facilitating discussions on various AI topics and fostering collaboration between academia, industry, and investment sectors [9][10]. Group 3: Notable Contributions and Events - The GAIR conferences have featured significant contributions from Microsoft scientists, addressing critical issues in AI, such as deep learning challenges and interdisciplinary integration [9]. - The upcoming eighth GAIR conference is scheduled for December 12-13, 2025, in Shenzhen, continuing the tradition of fostering innovative ideas and collaborations in the AI field [10].

Core Insights - The article discusses the introduction of a new framework called FedVLR, which addresses the challenges of multimodal integration in federated learning environments while ensuring data privacy [2][3][19]. Multimodal Integration Challenges - Current recommendation systems utilize multimodal information, such as images and text, but face difficulties in federated learning due to privacy concerns [2][5]. - Existing federated recommendation methods either sacrifice multimodal processing for privacy or apply a one-size-fits-all approach, which does not account for individual user preferences [2][5]. FedVLR Framework - The FedVLR framework redefines the decision-making flow for multimodal integration by offloading heavy computation to the server while allowing users to control how they view the data through a lightweight routing mechanism [3][19]. - It employs a two-layer fusion mechanism that decouples feature extraction from preference integration [8][19]. Server-Side Processing - The first layer involves server-side "multi-view pre-fusion," where the server processes data using powerful pre-trained models to create a set of candidate fusion views without burdening client devices [9][10]. - This approach ensures that the server prepares various "semi-finished" views that contain high-quality content understanding [10]. Client-Side Personalization - The second layer focuses on client-side "personalized refinement," utilizing a lightweight local mixture of experts (MoE) routing mechanism to dynamically compute personalized weights based on user interaction history [11][12]. - This process occurs entirely on the client side, ensuring that user preference data remains on the device [12]. Performance and Versatility - FedVLR is designed to be a pluggable layer that can integrate seamlessly with existing federated recommendation frameworks like FedAvg and FedNCF, without increasing communication overhead [16]. - The framework demonstrates model-agnostic capabilities, allowing it to enhance various baseline models significantly [26]. Experimental Results - The framework has been rigorously tested on public datasets across e-commerce and multimedia domains, showing substantial and stable improvements in core recommendation metrics like NDCG and HR [26]. - Notably, FedVLR performs exceptionally well in sparse data scenarios, effectively leveraging limited local data to understand item content [26]. Conclusion - FedVLR not only enhances recommendation systems but also provides a valuable paradigm for implementing federated foundational models, addressing the challenge of utilizing large cloud models while maintaining data privacy [19].