"Deep learning framework is the operating system of the smart times. Baidu fly paddle deep learning platform is to promote the rapid development of artificial intelligence, intelligent process to speed up China's industrialization." June 28, Baidu chief technology officer Wang Haifeng invited to attend the 2019 China International Software Fair, delivered a speech entitled "flying paddle deep learning platform to accelerate industrial intelligent" speech. Wang Haifeng shared thinking deep learning technology to promote the development of artificial intelligence in speech, describes the advantages of intelligent industry and boost the results of Baidu fly paddle (PaddlePaddle) deep learning platform.
Wang Haifeng said that so far three human experience of the industrial revolution, respectively, mechanical technology, electrical technology and information technology as the core driving force, and these techniques have shown a strong versatility. Currently, we are in a new round of technological revolution and industrial revolution tide of artificial intelligence at the core driving force, the human society into artificial intelligence is smart times.
Over the past 60 years time, the development of artificial intelligence has undergone artificial rules, machine learning and deep learning three typical stages. To OCR and natural language processing technology, for example, the depth of learning versatility advantages, making the effect of significantly improved, widely used in various industries.
Depth learning techniques are widely used, thanks to deep learning framework. Wang Haifeng pointed out that in the era of smart, deep learning framework play a connecting role, then under the chip, deck applications, great value, is the "era of intelligent operating system."
Scene, Wang Haifeng introduced China's first and currently the only open source open, fully functional depth learning platform - Baidu fly paddle (PaddlePaddle). Fly paddle core framework, including development, training and forecasting, as well as covering the recommendation, vision, natural language processing, speech, etc., rich model library. At the same time, flying paddles provided, including transfer learning, reinforcement learning, automated network design, elastic compute depth learning, neural networks, including a map tool components. In the service platform level, it provides a zero-based customized training and service platform EasyDL, one-stop training AI Studio development platform and end computing model generation platform EasyEdge and so on. Fly paddle to complete the framework, tools and services to help the majority of developers and businesses with tools, platform depth study of ways to further reduce the threshold for application, accelerate the transformation of industrial intelligence.
飞桨已开源70多个经过真实业务场景验证的官方模型，涵盖视觉、NLP、推荐等 AI核心技术领域，成为官方支持模型最多的深度学习平台。例如，面向工业应用的中文NLP工具集PaddleNLP，在应用任务层提供了中文词法分析、文本情感分析、短文语义匹配等能力，在基础网络层则纳入了当前业内效果最好的中文语义表示模型——ERNIE，大幅增强了模型语义表示能力，达到工业级的应用效果；而前不久刚刚发布的视频识别工具集，则覆盖了主流实用的序列建模算法与端到端视频识别模型，为开发者提供视频分类、视频定位等一系列应用技术方案。同时，飞桨具备适用大规模数据场景的分布式训练能力，从优化算力的角度为深度学习的应用提供支撑。
据悉，针对大规模稀疏特征，飞桨设计并开放了大规模稀疏参数服务器，适用于超大规模数据、海量特征和自膨胀、高频率模型迭代的业务场景。同时，针对工业级数据处理，飞桨开放了一系列数据处理技术服务于工业应用。
开发和训练后，将模型部署到各种应用场景下是非常关键的一步。飞桨提供包括底层硬件、推理引擎、多种程序设计语言、方案与服务工具在内的端到端全流程部署方案，已经在助力产业开发者实现应用的快速落地。
灵活、高效、易用是产业开发者应用深度学习技术进行开发的诉求。以往，传统神经网络的结构设计是由人根据经验设计，并不断地进行调参训练获得最优结果，这个过程较为复杂和费时费力。而飞桨开源的AutoDL Design自动化网络结构设计，是用深度学习设计深度学习，目前已经全面超过人类专家设计的网络效果，且自动化程度更高，效果更好。而针对没有深度学习技术基础的开发者，百度飞桨提供了定制化训练和服务平台EasyDL，让他们可以基于自身业务需求和数据，快速训练定制化AI模型。
Wang Haifeng automatic extraction farming land, for example, describes the practical application of fly paddle. Crop land paddle fly recognition based on the image data on the basis of remote sensing to achieve a 80% accuracy rate Lot extraction, artificial lift efficiency relative to the number of times, can be efficiently assist crop growth, disaster monitoring, and the like crop production, a significant reduction manpower, there is great significance to agricultural production, is the real depth of learning technology with the country.
At this stage, the depth of learning technology development breakthroughs, accelerate the integration of the industry, continued to enhance the commercial potential of value-added industries and accelerate industrial intelligence. Wang Haifeng said, "Baidu fly paddle deep learning platform will join China's vast number of developers together to promote China's development of intelligent industry to make more contributions."
ps: Finally, I recommend a GPU welfare - Tesla count V100 free power ! With PaddleHub situ model allows takeoff ~ ~ Application Fanger Wei code scan code