生成式人工智能2025年专利景观

Generative Artificial Intelligence 2025 Patent Landscape Table of Contents About Questel....................................................................................................................................3Executive Summary ......................................................................................................................4I.Introduction ............................................................................................................................... 6II.Methodology........................................................................................................................ 81.Data source & Search strategies................................................................................. 82.Taxonomy ................................................................................................................................... 9III.Deep learning global patent landscape ........................................................ 10IV.Focus on Multimodal AI / Digital Humans / Intelligent Agents....141.Introduction ..............................................................................................................................142.Multimodal AI...........................................................................................................................153.Intelligent Agents................................................................................................................. 174.Digital Humans........................................................................................................................18 About Questel Questel is a true end-to-end intellectual property solutions provider to more than20,000 clients and 1,5M users across 30 countries. We offer a comprehensive software Questel also provides services throughout the IP lifecycle, including prior artsearches, patent drafting, international filing, translation, and renewals. These Questel’s mission is to allow innovation to be developed in an efficient, secure, andsustainable way. Behind this mission, Questel considers that Corporate Social More content is available online, please consult: www.questel.com/resources-hub/ Executive Summary Deep learning continues to be one of the most dynamic areas of technological innovation,with patent activity showing sustained and accelerating growth. Building on our previousstudy on Deep Learning and Large Language Models, this report analyzes the next wave of The patent landscape reveals that a small number of global technology leaders are shapingthis transition by combining foundation models, agentic capabilities, and human-like GOOGLE positions itself as a core technology leader through its Gemini family ofmodels. Designed as natively multimodal, Gemini integrates text, vision, audio, and BAIDU stands out as the most vertically integrated player. With the ERNIE multimodalengine, GenFlow and AgentBuilder for intelligent agents, and a rapidly expanding NVIDIA dominates the Digital Humans landscape. While not positioning itself as ageneral-purpose AI assistant provider, NVIDIA supplies the essential infrastructure, MICROSOFT adopts an enterprise-first approach that connects multimodal AI andintelligent agents directly to business workflows. Leveraging both its partnership cloud, and enterprise software. Its patent filings show a balanced strategy, combiningsolid portfolio size with a high proportion of international families, reflecting global IBM emerges as a key leader in Intelligent Agents, supported by its watsonx.aiplatformand Granite foundation model family.IBM’s strategy focuses on I.Introduction Deep Learning represents a core technological layer within artificial intelligence, based onmulti-layered artificial neural networks that learn hierarchical representations from largedatasets and excel in perception tasks such as image recognition, speech recognition, andnatural language processing. Within this technological layer, Generative AI (GenAI) emerged In contrast to these foundational AI layers, several domains have developed as applications •MultimodalAI refers to systems that can process,understand,and generateinformation across multiple modalities or data types text, images, audio, video, outputs that span modalities (e.g., a captioned image, a narrated video, or asynchronized audio-visual response). This multimodal capability enables richer •Intelligent Agents are autonomous or semi-autonomous AI systems that perceivetheir environment via sensors (which could include vision, audio, or other sensordata), make decisions based on given objectives, and take actions to achieve specificgoals. Built on deep learning, multimodal understanding, and generative or reasoning •Finally, Digital Humans are AI-powered virtual representations of human beings,combining computergraphics,animation,natural language processing,speechsynthesis,and behavioral AI to create realistic human avatars cap