本报讯 （记者贾丽）近日，"异构融合创新未来，智算赋能海洋科技"主题研讨会在青岛举行，异构智 算产业生态联盟携手英特尔等生态合作伙伴等走进青岛国实科技集团有限公司（以下简称"青岛国实集 团"），聚焦GPU+CPU混合架构在深海勘探、海洋碳汇监测等场景的创新应用进行了分享交流。 AI赋能海洋 打造蓝色科技创新引擎 全国政协委员、中国科学院计算技术研究所研究员张云泉表示，人工智能大模型技术正经历从实验室研 究到产业落地的关键转型期，并引领着算力行业向更高层次、更广领域、更深融合的方向发展。 作为海洋科技领域的国家级平台，崂山实验室正以海洋科技与异构智算相融合，为海洋强国建设提供强 有力战略科技支撑。崂山实验室副主任、研究员吴能友表示："实验室承载着国家战略科技力量的重 任，希望能与合作伙伴一起推动关键核心技术协同攻关，加强颠覆性技术和前沿技术的研究，推动海洋 产业高质量发展。" 大模型技术的快速发展为海洋人工智能提供了坚实的基础。作为青岛海洋科技产业群新的生力军，青岛 国实集团紧紧围绕海洋科技产业，积极推进科技成果转化和产业化。 算力、模型和应用融合发展 面向海洋观测、深海勘探等领域，人工智能大模型正深刻重塑海洋 ...

Group 1 - The event "Heterogeneous Intelligent Computing, Shaping the Future" focused on the deep integration of heterogeneous intelligent computing and AI large models, highlighting the advancements in AI technology and its applications across various sectors [1][3] - The emergence of DeepSeek represents a significant technological breakthrough in AI, particularly in low-computing training and high-efficiency model optimization, which is expected to enhance AI's penetration in fields such as healthcare, education, automotive, and finance [3][5] - The "Digital Life" major scientific plan initiated by Peking University aims to leverage multimodal cross-scale biomedical imaging facilities to visualize and measure life phenomena and disease mechanisms, addressing complex life science challenges [3][5] Group 2 - The integration of AI large models with life sciences is transforming drug development, with a shift towards intelligent pharmaceutical research that emphasizes precision, efficiency, and cost-effectiveness [5][6] - Lenovo's AI infrastructure strategy, particularly the Wanquan heterogeneous intelligent computing platform, aims to address the challenges of intelligent computing with advanced liquid cooling technology and high-precision interconnects, promoting industrial upgrades [6][7] - The Wanquan platform is designed to manage heterogeneous computing clusters efficiently, enabling seamless access to integrated, stable, and intelligent computing resources, and can support the full-speed operation of the DeepSeek R1 large model [6][7]