百度研究院：2020年10大人工智能科技趨勢

近日，百度研究院發(fā)布了一份關(guān)于2020年人工智能科技趨勢預(yù)測的報(bào)告，報(bào)告從十個(gè)角度對2020年AI的主要發(fā)展趨勢做了闡述。

以下是十大預(yù)測趨勢的詳細(xì)解讀：

趨勢一：AI 技術(shù)已發(fā)展到可大規(guī)模產(chǎn)業(yè)化階段，2020年將出現(xiàn)多家AI工廠

AI技術(shù)以及各類商業(yè)解決方案已日臻成熟，并快速進(jìn)入產(chǎn)業(yè)化階段。隨著全球科技巨頭對AI技術(shù)的持續(xù)投入，到2020年，全球范圍內(nèi)將出現(xiàn)多家人工智能模型與數(shù)據(jù)工廠，從而大規(guī)模推動人工智能技術(shù)和相關(guān)的商業(yè)解決方案更新產(chǎn)業(yè)。例如客服行業(yè)的AI解決方案將可以大規(guī)模復(fù)制運(yùn)用到金融、電商、教育等諸多行業(yè)中。

(英文原文，下同：The increasingly mature AI technology and all types of associated business solutions are rapidly entering the stage of "industrialization". With the continuous investment global technology giants pumped into AI technology, there will be many factories of AI models and data emerging in 2020, facilitating AI technology and associated commercial solutions on a large scale to update industries. For example, AI solutions in the customer service industry can be copied and applied to finance, e-commerce, education and other industries on a large scale. )

趨勢二：2020年將會是AI芯片落地的關(guān)鍵年

近幾年，AI芯片逐漸達(dá)到了可用的狀態(tài)，2020年將會是AI芯片大規(guī)模落地應(yīng)用的關(guān)鍵一年。端側(cè)AI芯片將更加低成本、專業(yè)化、解決方案集成化。同時(shí)，神經(jīng)網(wǎng)絡(luò)處理單元(NPU)將成為下一代端側(cè)通用CPU芯片的基本模塊，未來越來越多的端側(cè)CPU芯片將會以深度學(xué)習(xí)為核心進(jìn)行全新的芯片規(guī)劃。除了芯片以外，AI還將重新定義計(jì)算機(jī)體系架構(gòu)，支持人工智能訓(xùn)練和推理，成為異構(gòu)設(shè)計(jì)架構(gòu)的新思路。

(In recent years, AI chips have gradually reached a usable state, and 2020 will be a critical year for the large-scale implementation of AI chips. AI chips on the edge will be more low-cost, specialized and seamlessly integrated into downstream solutions. At the same time, the neural processing unit (NPU) will become the basic module of the next-generation edge-based general-purpose CPU chips. In the future, more and more device-based CPU chips will integrate deep learning framework as the core to their designs. In addition to chips, AI will redefine the computer architecture and support AI training and inference as a new idea of heterogeneous design architecture. )

趨勢三：深度學(xué)習(xí)深入滲透產(chǎn)業(yè)，并大規(guī)模應(yīng)用

深度學(xué)習(xí)是人工智能領(lǐng)域最重要，也是被產(chǎn)業(yè)界證明最有效的技術(shù)。以深度學(xué)習(xí)框架為核心的開源深度學(xué)習(xí)平臺大大降低了人工智能技術(shù)的開發(fā)門檻，有效提高了人工智能應(yīng)用的質(zhì)量和效率。2020年，深度學(xué)習(xí)將大規(guī)模應(yīng)用于多個(gè)行業(yè)，實(shí)施創(chuàng)新，加快轉(zhuǎn)型升級。

(Deep learning is the most important and effective technology in the field of artificial intelligence. At the core of open-sourced deep learning platforms is the deep learning framework, which greatly lowers the development threshold of AI technology, and effectively improves the quality and efficiency of AI applications. In 2020, deep learning will be applied across many industries at scale to implement innovation and accelerate transformation and upgrading. )

趨勢四：AutoML將大大降低機(jī)器學(xué)習(xí)的門檻

AutoML的快速發(fā)展將大大降低機(jī)器學(xué)習(xí)的門檻，擴(kuò)大AI應(yīng)用的普及率。AutoML將能夠把傳統(tǒng)機(jī)器學(xué)習(xí)中的迭代過程綜合在一起，構(gòu)建一個(gè)自動化的過程。研究人員只需輸入元知識(如卷積運(yùn)算、問題描述等)，算法就可以自動選擇合適的數(shù)據(jù)、優(yōu)化模型結(jié)構(gòu)和配置、自動地訓(xùn)練模型，并將其部署到不同的設(shè)備上。

(AutoML will be able to integrate the iterative process in traditional machine learning and build an automatic process. Researchers only need to input meta-knowledge (such as convolution operations, problem descriptions, etc.), the algorithm can automatically select the appropriate data, optimize the model structure and configuration, train the model, and deploy it on different devices. The rapid development of AutoML will greatly lower the threshold of machine learning and increase the popularity of AI applications. )

趨勢五: 多模態(tài)深度語義理解進(jìn)一步成熟，并得到更廣泛應(yīng)用

多模態(tài)深度語義理解以聲音、圖像、文本等不同模態(tài)的信息為輸入，融合感知和認(rèn)知技術(shù)，實(shí)現(xiàn)對信息的多維度深層次理解。隨著計(jì)算視覺、語音、自然語言理解和知識圖譜等技術(shù)的快速發(fā)展和大規(guī)模應(yīng)用，多模態(tài)深度語義理解逐漸成熟，應(yīng)用場景更加廣闊。結(jié)合AI芯片，將廣泛應(yīng)用于智能家居、金融、安防、教育、醫(yī)療等行業(yè)。

(Multimodal deep semantic understanding takes the information of different models such as voice, image, and text as input, and integrates perception and cognition technologies to achieve a multi-dimensional deep understanding of information. With the rapid development and large-scale application of computing vision, speech, natural language understanding, and knowledge graph, multimodal deep semantic understanding is gradually mature, which leads to a broader application scenario. Combined with AI chips, it will be widely used smart home, finance, security, education, healthcare, and other industries. )

趨勢六：自然語言處理技術(shù)將與知識深度融合，面向通用自然語言理解的計(jì)算平臺得到廣泛應(yīng)用

隨著大規(guī)模語言模型預(yù)訓(xùn)練技術(shù)的出現(xiàn)和發(fā)展，通用自然語言理解能力有了極大地提高?；诤Ａ课谋緮?shù)據(jù)的語義表示預(yù)處理技術(shù)將與領(lǐng)域知識深度融合，不斷提高自動答疑、情感分析、閱讀理解、推理、信息提取等自然語言處理任務(wù)的有效性。集合超大規(guī)模算力、豐富領(lǐng)域數(shù)據(jù)、預(yù)訓(xùn)練模型和完善研發(fā)工具的通用自然語言理解計(jì)算平臺將逐漸成熟，并在互聯(lián)網(wǎng)、醫(yī)療、法律、金融等領(lǐng)域得到廣泛應(yīng)用。

(With the emergence and development of pre-training large-scale language model, the technology of general natural language understanding has been greatly improved. Semantic representation pre-training technology based on massive text data will be deeply integrated with domain knowledge to continuously improve the effectiveness of natural language processing tasks such as automatic question answering, emotional analysis, reading comprehension, reasoning, information extraction, etc. The general natural language understanding the computing platform, which integrates large-scale computing power, rich domain data, pre-training model and improved R&D tools, will be gradually improved and widely used in the internet, healthcare, legal, financial and other fields. )

趨勢七：物聯(lián)網(wǎng)將在邊界、維度和場景三個(gè)領(lǐng)域形成突破

隨著5G和邊緣計(jì)算的發(fā)展，算力將不再局限于云計(jì)算中心，向萬物蔓延，會產(chǎn)生一個(gè)泛分布式計(jì)算平臺。同時(shí)，對時(shí)間和空間這兩個(gè)物理世界最重要維度的洞察，將成為新一代物聯(lián)網(wǎng)平臺的基本能力。這也將推動物聯(lián)網(wǎng)與能源、電力、工業(yè)、物流、醫(yī)療、智能城市等更多場景發(fā)生融合，創(chuàng)造出更大的價(jià)值。

(With the development of 5G and edge computing, computing power will not be limited to cloud computing centers, expanding to everything and building a distributed computing platform. At the same time, the insight into time and space, the two most important dimensions of the physical world, will become the basic capabilities of the new-generation IoT platforms. This will promote the integration of IoT with more scenarios such as energy, power, industry, logistics, medical treatment, and intelligent city, and create greater value.)

趨勢八:智能交通將加速在園區(qū)、城市等多樣化場景中落地

自動駕駛的發(fā)展正在趨于理性，未來幾年市場對智能駕駛的發(fā)展也會更加有信心。2020年，自動駕駛汽車將被應(yīng)用于物流快遞、公共交通、封閉道路等不同場景。同時(shí)，V2X(vehicle to everything)技術(shù)啟動規(guī)模化部署和應(yīng)用，這使得車輛和道路形成一個(gè)廣泛的聯(lián)系，進(jìn)一步推動智能車路協(xié)同技術(shù)的實(shí)現(xiàn)，智能交通加速在園區(qū)、城市、高速等多樣化場景中落地。

(The development of autonomous vehicles is becoming more rational, and the market will be more confident in the development of intelligent driving in the next few years. In 2020, more autonomous vehicles will be applied to different scenarios such as logistics, public transport, geofenced areas and so on. At the same time, V2X (vehicle to everything) technology is ready for large-scale deployment and application, which makes vehicles and roads form a wide range of connections, further promoting the realization of Intelligent Vehicle Infrastructure Cooperative Systems (IVICS), and accelerating the implementation of intelligent transportation in parks, cities, expressways and other scenarios. )

趨勢九：區(qū)塊鏈技術(shù)將以更加務(wù)實(shí)的姿態(tài)融入更多場景

隨著區(qū)塊鏈技術(shù)與人工智能、大數(shù)據(jù)、物聯(lián)網(wǎng)以及邊緣計(jì)算的深度融合，數(shù)據(jù)與資產(chǎn)的線上線下映射問題將逐一解決。圍繞區(qū)塊鏈構(gòu)建的數(shù)據(jù)確權(quán)、數(shù)據(jù)使用，數(shù)據(jù)流通和交換等解決方案，將在各行各業(yè)發(fā)揮巨大的作用。例如，在電商領(lǐng)域，可保證商品全流程數(shù)據(jù)的真實(shí)性;在供應(yīng)鏈領(lǐng)域，可保證全流程數(shù)據(jù)的公開和透明，以及企業(yè)之間的安全交換;在政務(wù)領(lǐng)域，可以實(shí)現(xiàn)政府?dāng)?shù)據(jù)的打通、電子證書的實(shí)現(xiàn)等。

(With the in-depth integration of blockchain technology with AI, big data, IoT and edge computing, the problems concerning the online and offline mapping of data and assets will be solved one by one. Solutions such as data authorization, data use, data circulation and exchange built around blockchain will play a huge role among people from all walks of life. For example, in e-commerce, blockchain can ensure the authenticity of the whole process data of goods; in supply chain, it can ensure the openness and transparency of the whole process data, as well as the safe exchange between enterprises; in government affairs, it can achieve the opening of government data, the realization of electronic certificates and so on. )

趨勢十：量子計(jì)算將迎來新一輪爆發(fā)，為AI與云計(jì)算注入新活力

隨著量子霸權(quán)的成功展示，量子計(jì)算將在2020年迎來新一輪爆發(fā)。量子硬件方面，可編程的中等規(guī)模有噪量子設(shè)備的性能會得到進(jìn)一步提升，并具備糾錯能力。具有一定實(shí)用價(jià)值的量子算法將能夠在其上運(yùn)行，量子人工智能的應(yīng)用將得到極大的發(fā)展。

量子軟件方面，高質(zhì)量的量子計(jì)算平臺和軟件將會出現(xiàn)，并與AI和云計(jì)算技術(shù)深度融合。此外，隨著量子計(jì)算生態(tài)產(chǎn)業(yè)鏈的初步形成，量子計(jì)算必將在更多應(yīng)用領(lǐng)域受到更多的關(guān)注。越來越多的行業(yè)巨頭陸續(xù)投入研發(fā)資源進(jìn)行戰(zhàn)略布局，這將給未來的人工智能和云計(jì)算領(lǐng)域帶來新的面貌。

(With the successful demonstration of quantum hegemony, quantum computing will usher in a new round of explosive growth in 2020. In terms of quantum hardware, the performance of programmable medium-sized noisy quantum devices will be further improved and have the ability of error correction. Quantum algorithms with certain practical value will be able to run on them, and the application of quantum artificial intelligence will be greatly developed. In terms of quantum software, high-quality quantum computing platforms and software will emerge and be deeply integrated with AI and cloud computing technologies. In addition, with the emergence of the quantum computing industry chain, quantum computing will surely garner more attention in more application fields. More and more industry giants have invested in R&D resources for strategic layout, which has the opportunity to bring a new face to the future AI and cloud computing fields. )