科大讯飞联合创始人兼高级副总裁江涛:AI连接万物,共创新时代 | 2019 WISE 超级进化者大会


On July 9-10, 36氪 held the “2019WISE Super Evolutionist” conference in Beijing and Shanghai. The event has seven conference venues, focusing on the path of enterprise development and change, the grasp of industry trends, the advancement and transformation of the retail industry, and the trillions of enterprises. Topics such as the rise of the service market, industrial innovation opportunities, globalization trends and the outbreak of differentiated demand invite more than 100 industry leaders to focus on the rise of super-evolutionists who lead the industry. In the era of Internet of Everything, artificial intelligence has achieved rapid development, and technologies such as speech synthesis and recognition have also landed on many scenes such as justice, education, medical care, and family. In the future, there are other applications that can be industrialized on a large scale. The continuous iteration from algorithm to application has become the most concerned issue, and it has become the biggest industrial opportunity in the field of artificial intelligence in the future. In this regard, Jiang Tao, co-founder and senior vice president of Keda Xunfei, believes that on the one hand, the machine should be able to understand, think, and move from cognitive intelligence to cognitive intelligence. On the other hand, artificial intelligence should not only achieve technological breakthroughs, but also create a product that has a large amount of coverage and can be commercially successful and influential, so that users are really willing to pay for the purchase and “vote” for your product. The following is a guest speech: everyone developers, entrepreneurs, good morning everyone! Keda Xunfei is a listed company developed by a group of Chinese university students. In 1998, the team members of Keda Xunfei were also a group of researchers in the laboratory of the University of Science and Technology of China. In 1999, this small company was established. Today we see this theme “super evolver”, we are still quite touched. In 2008, the University of Science and Technology News was listed, with a market capitalization of 3.3 billion. Ten years after the listing, the scale has been further expanded, and we are also facing the huge industrial opportunities in the era of artificial intelligence. Today I want to share with you our views on industrial opportunities in the era of artificial intelligence. From the perspective of the University of Science and Technology, what stage has artificial intelligence been developed? First of all, I would like to review with you the work that the University of Science and Technology has done over the past decade and on artificial intelligence. In 2010, two companies around the world based on deep learning algorithms made the first breakthrough in speech recognition. One is Google, doing continuous speech recognition in English. The other is Keda Xunfei. On October 28, 2010, at the Shangri-La Hotel, Beijing, the world’s first continuous Chinese speech recognition system was released. This system is also leading the world. It took us a few years to further deepen the theoretical framework of deep learning and found that deep learning of this system is not only a solution to the problem that the machine can hear, but also can be extended to understand and think. A lot of regular and repetitive mental work, get enough process data in a specific task, there are enough field experts to train us, the machine can continue to approach the expert level in this field on a specific task. In 2014, we opened a big conference and officially launched the “Xunfei Super Brain” project. The problem to be solved is to let the machine understand and understand, from perception to intelligence, to perception. intelligent. Keda Xunfei is trying hard to explore from the audible and understandable thinking, which applications can be scaled up, and can achieve continuous iteration from algorithm to application. In the past few years of exploration, in 2018, the University of Science and Technology has already ran a number of very good applications on the track including education, medical treatment and justice. At the meeting of our company, Chairman Liu Qingfeng told everyone that the strategy of the artificial intelligence of the University of Science and Technology has reached the 2.0 stage, which is the bonus period. Artificial intelligence is not only to achieve technological breakthroughs, but also to create a large and wide range of products that can be commercially successful and influential. Let’s go back and see what stage of the core technology of artificial intelligence has developed. Let’s first look at speech synthesis technology. This year, the Central Radio and Television Headquarters and the University of Science and Technology News, based on our technology, made the world’s first multilingual virtual journalist “Unicom”. We filmed the image of the host and recorded a short video that allowed the machine to speak the mainstream language of the world as a host. This can be used not only in CCTV, but also in media partners like 36氪, which can make your content more interesting and more personalized.语音识别这几年取得了突飞猛进的发展,但随着万物互联时代的到来,语音识别的要求也越来越高了。在万物互联时代,很多设备离我们比较远,比如智能家居。如果你要走到设备的开关面前再说指令就没有意义了,还不如直接摁一下开关。智能家居,就是要在远距离情况下,能控制家用电器。在高噪音的情况下,比如一边开车,一边语音命令控制。在远距离、高噪声、多人说话的情况下,语音智能还能不能做到,这才能体现出万物互联的价值。去年,在号称史上最难的语音识别比赛CHIME-5上(一个在厨房、起居室等各种场景下进行的高难度比赛),科大讯飞作为中国代表击败了美国同行,拿到了全球最好的水平,在所有的四个主要指标上都囊括了全球第一。基于这些技术的突破,人工智能从去年到今年在大规模的进行产业化落地。科大讯飞也发布了一系列围绕办公场景的产品,比如转写机,现在有一些基金公司在购买使用。它是一个小箱子,一打开可以实现会议记录。会议开完以后,生成一个二维码,所有在场的人一扫二维码,从PPT到演讲者说的话、从语音到文字,都可以直接分享查看。我们还有智能办公本,大家开会中说的语音内容也可以变成结构化的文字。除此之外,会议记录系统、智能录音笔,都在618取得了很好的售卖成绩。我们是把办公场景产生的声音和手写的文字,变得可保存、可检索、可统计、可分析,极大提高办公效率。我相信这个场景未来有很大的空间,这是人工智能实实在在落地的、已经达到的规模化应用。当然光讲技术突破没有用,还要看用户是不是拿他的钱来为你的产品“投票”。刚刚过去的618,在京东、天猫这些平台上,科大讯飞围绕办公场景做的一系列产品,都得到了市场的高度认同,比如“机器翻译”。科大讯飞去年的机器翻译推动实际规模数据的统计对比结果,已经远远超过六级考试的大学生水平。而基于这项技术开发的讯飞翻译机,也得到了市场的认可,已经覆盖了200个国家和地区,在今年618众多翻译机品类中遥遥领先,得到的用户的广泛认可。刚刚说了“能听会说”的规模化应用,从2014年科大讯飞启动“讯飞超脑”之后,科大讯飞重点加码的领域,是从能听会说到能理解会思考。围绕很多脑力劳动者,也就是产业互联网中的需求,除了语音之外,进一步到了认知层面。这些脑力劳动者在特定任务上的表现,机器也在不断学习。这里有一项基础技术叫机器阅读理解技术,也就是机器对于人类自然语言描述的文章能读懂多少。在“讯飞超脑”项目支持下,这方面的研发加大了力度。去年1月22号的SQuAD上,科大讯飞首次在F1指标上,超过美国大学生水平。这是斯坦福举办的机器阅读理解的挑战赛,谷歌、微软、IBM、Facebook等全球主要的科技巨头都参加了。尤其是微软,这几年持续加大对人工智能的重视,且微软过去几年里都在这项比赛中保持了全球领先。但是2018年1月份,我们超过了微软。目前在精确匹配指标上,全球有两家公司超过了美国大学生水平,一是家微软,一家是科大讯飞,而且我们得分比微软还高。今年3月20号,我们用全新的算法,再次刷新了记录。在SQuAD 2.0比赛中,精确匹配和模糊匹配,我们都超过了美国大学生的水平,这是机器阅读理解领域非常有里程碑意义的测试。 机器阅读理解达到了这样的水平,在应用上有什么变化呢?教育领域是个例子。在前年一些典型的比赛比如高考、研究生考试上,机器在批改作文的主观题上,已经达到了人类专家的水平。Whether the machine is small, medium or large, it has obvious advantages over the human marking teacher, and it is more accurate and more stable. Based on these technological breakthroughs, we can systematically analyze the content written by children on paper through artificial intelligence, and then collect the process data of the children’s homework and exams, analyze the weak links of the children, and provide him with personalized guidance programs. In addition to this, there are applications in the medical field. Keda Xunfei is the only technical system in the world that has passed the National Physician Qualification Examination. The national doctor’s qualification examination scored 600 points, the University of Science and Technology’s system ran 456 points, more than 96.3% of the 530,000 examinations in 2017, the specific clinical problem machine can analyze and give solutions. Last year, we put this technology into a product and put it into use. By the beginning of this year, there were 13,000 cases of auxiliary medical treatment in real scenes every day, covering more than 900 common diseases. This system is not for the top doctors, but for doctors in the grassroots hospital community and county hospitals to reduce their missed diagnosis and misdiagnosis. 95% of common diseases, the machine can provide advice, and the accuracy of diagnosis is also increasing, currently reaching 97%. As the size of the data continues to increase, the accuracy of the future will continue to increase. Of course, artificial intelligence is a great historical process, not three or five years. The core technology of artificial intelligence is based on algorithms, including continuous training on small samples and unsupervised scenes. How to combine with brain science and brain science, there is still a lot of work to be done. In the application mode, artificial intelligence has a lot of work to be explored. From the core technologies of perceived intelligence to cognitive intelligence, we have packaged it into the open platform of the University of Science and Technology, which is open to all developers. There are already 1 million developers on this platform, and there are many entrepreneurs present. I hope everyone can come to the open artificial intelligence platform of Keda Xunfei. You don’t have to worry about how complex the underlying technology implementation is. If you do good things in a familiar field, hand over the complex artificial intelligence algorithms to the University of Science and Technology. Through this platform, we hope to help everyone achieve success together. .

Tags : Internet entrepreneurship互联网创业项目创业