Quality Alchemist (品質煉金術師): CAAI Chinese Congress on Artificial Intelligence 2020 (中国人工智能大会)

2020年8月29日星期六

CAAI Chinese Congress on Artificial Intelligence 2020 (中国人工智能大会) - Day 1

The 6th Chinese Congress on Artificial Intelligence (CCAI2020) which was led by Chinese Association for Artificial Intelligence (CAAI), was held on Aug 29-30, 2020, in Nanjing, China. Since COVID-19 affected globally, the Congress was run both online and offline. The frame of congress named “Intellectual Know for Everything (智周万物)”. I was committee member of the CAAI Extenics Professional Committee (中國人工智能學會可拓學專業委員會) attended the congress online in this time.

Before the congress, a video was played to introduce the CCAI 2020 (https://youtu.be/Yqk9YSY0rH8).

Day 1 (29 Aug 2020): (Opening Session)

In the beginning, Prof. Zhou Zhihua (周志华 – CAAI副理事长, 南京大学计算机系主任, 人工智能学院院长) chaired this session and he introduced today’s guest speakers.

Prof. Dai Qionghai (戴琼海 - CAAI理事长，中国工程院院士，清华大学信息学院院长) gave guest speech. Prof. Dai said CCAI 2020 as one of the largest official AI conferences in China, CCAI has been held annually to promote the advancement of artificial intelligence globally. He appreciated all co-organizers and supporting organizations.

Mr. Xu Chunsheng (徐春生 – 江苏省科学技术协会副主席) gave a guest speech and he said Jiangsu Province enhanced the development on AI including industry and education. He expected they could achieve the AI demonstration province in 2025.

Mr. Liang Aiguo (梁爱国 - 南京市建邺区副书记) gave a guest speech. He welcome all guests in Jianye District, Nanjing. CCAI as AI platform would enhance AI development in Jianye District.

Then Mr. Yang Bo (杨波 – 南京市建邺区政府副区长, 建邺高新区党工委书记) gave an introduction speech for investment and environment in Jianye District, Nanjing. Jiangsu Nanjing Eco Hi-Tech Island Economy Development Zone (江苏南京生态科技岛) was introduced at the end.

After that Prof. Ye Jieping (叶杰平 – 滴滴出行副总裁, 滴滴AI Labs 负责人, 密西根大学教授) chaired the keynotes session.

The first speaker was Prof. Yoshua Bengio (加拿大蒙特利尔大学教授，2018年图灵奖获得者，蒙特利尔学习算法研究) and his topic entitled “Deep Learning: from System 1 to System 2”. Firstly, Prof. Bengio briefed the neural net approach to AI including Brain-inspired. He said not copy brain but inspired by brain to AI.

Then he quoted the concept of System 1 and System 2 in the book “Thinking, Fast and Slow” to explain the cognition.

And then he continued to explain all items inspired by human cognition one by one. Prof. Bengio added consciousness from humans reporting such as high-level representation (e.g. language) and high-level concepts (e.g. tie system 1&2) and needed to understand grounded language learning (e.g. BabyAI).

Finally, Prof. Bengio concluded we have a responsibility that ML going out of labs, into society and considered society impact.

The second speaker was Prof. Yang Qiang (杨强 - CAAI名誉副理事长，微众银行首席AI官) and his topic named “The latest development of Federated Learning” (联邦学习最新进展). He said AI power come from Big Data that increasing of Data caused error reducing and accuracy increasing every year.

Then Prof. Yang Qiang said many small data as an island which were not able to contribute for AI model. Therefore, Federated Learning could overcome this problem to consolidate those small data without losing their privacy information (e.g. GDPR violations). He added that Data did not move but model move!

And then he briefed the data privacy development in Shenzhen that those rules become strict.

After that Prof. Yang briefed Federated Learning had two types. One is Federated Averaging (Horizontal) that google employed.

The other one is Vertical Federated Learning that he proposed in 2018. Prof. Yang explained that two banks customers had different characteristics of those data respectively. They would like to cooperate and fulfill privacy regulations.

Lastly, Prof. Yang Qiang expected to form alliance of Federated Learning that different industries could cooperation to employ and share their data privately. The problem needed to solve is how to calculate the contribution and to distribute the benefit fairly.

The third speaker was Prof. Dai Qionghai (戴琼海 - CAAI理事长，中国工程院院士，清华大学信息学院院长) and his presentation entitled “Artificial Intelligence: Algorithm. Computing Power. Interaction” (人工智能：算法.算力.交互). In the beginning, he reviewed AI history and then explain the development of computing power.

He also quoted Neil C. Thompson et al. paper named “The Computational Limits of Deep Learning” that deep learning approached to the computational limit of chips. And showed the demand of computational power trend diagram.

After that he mentioned the Algorithm development and it inspired by brain study. That was the direction of future Algorithm development of AI.

Finally, Prof. Dai stated the interaction between AI and human and then cooperation with each other. He raised some examples such as in auto-driving, surgery, service robot, etc.

At the end, Prof. Dai concluded the future machine-human co-working harmony, cognition computing mechanism and testing, new carrier of computation would be happened.

In afternoon, there were four parallel session and I joined the session named “Intelligent Epidemic Prevention Forum” (智能防疫专题论坛) and then attended “Human-and-Machine Intelligence Forum” (人与机器智能专题论坛).

The first speaker in the session of “Intelligent Epidemic Prevention Forum” was Prof. Sun Fuchun (孙富春 - CAAI副理事长，清华大学计算机科学与技术系教授，国家杰青) and his presentation entitled “The development and prospect of intelligent medical robots” (智能医疗机器人的发展及展望). Firstly, Prof. Sun briefed the development history of medical robots.

Then Prof. Sun mentioned the core technology of medical robots that included motion safety and operation feasibility.

Finally, Prof. Sun expected the future medical robots had AI behavior included structure design, operation control, human-machine interaction, environment rebuilt and active learning.

The second speaker was Prof. Li Juanzi (李涓子 - 清华大学教授、人工智能研究院知识智能中心主任) and her topic named “Aminer Knowledge Epidemic Map (Aminer知识疫图)”. Firstly, Prof. Li briefed the AMiner Big Data mining and developed a knowledge service platform.

She said AMiner Knowledge Epidemic Map developed to monitor the global COVID-19 epidemic for risk assessment. Their risk index model included epidemic data, region population and area, as well as, region medical resource.

The third speaker was Dr. Zheng Yefeng (郑冶枫 - 腾讯天衍实验室主任，AIMBE Fellow) and his presentation entitled “Medical AI helps fighting the new Corona Pneumonia: Tencent Medical's Anti-epidemic Story” (医疗AI助力抗击新冠肺炎：腾讯医疗的抗疫故事). In the beginning, Dr. Zheng mentioned their laboratory established since 2018.

After that he briefed three technical difficulties included small sample learning, small different between Covid-19 pneumonia and common pneumonia, as well as, not enough power of model generalization.

Finally, they overcome all problems and launched Tencent Health. Dr. Zheng said their model based on D-SEIQ that changed from SEIR model to SEIQ model (where R is for recover and Q is to isolate.)

The forth speaker was Mr. Zhang Rong Guo (张荣国 -北京推想科技有限公司先进研究院院长) and his presentation topic named “The image is the outpost, AI is the sentinel – intelligent pneumonia screening and epidemic monitoring system” (影像为前哨，AI做哨兵—肺炎智能筛查及疫情监测系统). He employed CT and AI for screening COVID-19.

Mr. Guo cooperated with three hospital and employed their CT for AI training. Then compared with external valid data and got the very good accuracy. Therefore, AI + CT is used to screen for isolation and monitoring, PCR used for final confirmation.

The fifth speaker was Prof. Yang Bin (杨斌 -清华大学网络行为研究所副所长，清华大学精准医学研究院智慧健康中心主任) and his topic was “COVID-19 Intelligent Prevention and Control System” (新冠智能防控系统). He firstly said the situation in Wuhan was no bed, closely contact and cross-infection.

Then they proposed a solution through family individual protection, social district isolation, and hospital classified handling, as well as, social district isolation monitoring.

Finally, Prof. Yang showed their product and system which employed to help epidemic area.

The other parallel session named “Future Technology · Empowering Industry Upgrade Forum” (未来科技·赋能产业升级专题论坛) had not finished. The panel discussion named “How Big Data and AI technologies can Empower Industry Development” (大数据与人工智能技术如何赋能产业发展).

After that we went to another parallel session named “Human-and-Machine Intelligence Forum” (人与机器智能专题论坛).

The first speaker in the Smart City Session was Prof. Jiang Yi (蒋毅 - 中科院心理研究所研究员，国家杰青) and his topic entitled “Visual Perception Processing of Biological Movement” (生物运动的视知觉加工).

Finally, he summarized that “a genetic linkage between local Biological Movement (BM) processing and autistic traits” and “Opens up the possibility of treating the ability to process local BM information as a distinct hallmark of social cognition.”

The second speaker was Prof. Liu Jia (刘嘉 - 北京师范大学心理学部教授，国家杰青，长江学者特聘教授) and his topic named “Cognitive Neural Computing: From Representation to Cognition” (认知神经计算：从表征到认知). Firstly, Prof. Liu introduced GPT-3 (Generative Pre-training Transformer) which was a powerful AI using huge computation power but it cannot answer dumb question. Then Prof. Liu quoted Einstein statement “Any fool can know. The point is to understand.”

Then Prof. Liu discussed the intelligence that should be located in dynamic and open environment. He said different between human intelligence and AI was same as René Descartes (笛卡兒) statement “Even though such machines might do some things as well as we do them, or perhaps even better, they would inevitably fail in others, which would reveal they were acting not through understanding, but only from the disposition of their organs.”

Finally, he raised three points as taking home messages. They were similar representation in different systems to achieve same goal, prior experience as a deterministic role, and cognitive neuroscience provided powerful paradigms and tools.

The third speaker was Prof. Wu Si (吴思 - 北京大学信息科学技术学院教授，麦戈文脑科学所研究员) and his presentation was “Neural expression of Time” (时间的神经表达). Beginning, Prof. Wu briefed neural computational science was the bridge between brain science and AI.

And then Prof. Wu discussed time and neural computation in the temporal domain. He also mentioned some algorithm and said “Activating a hub neuron is difficult”. Hub Neurons trigger synchronous firing; Loop formed by low-degree neurons define the rhythm was showed.

The forth speaker was Prof. Cheng Xiuzhen (吴华强 -清华大学微纳电子系副主任、教授，清华大学微纳加工平台主任，北京市未来芯片技术高精尖创新中心副主任) and his presentation named “Brain-inspired Storage and Computation Integration and New Neural Network” (脑启发的存算一体与新型神经网络).

Prof. Cheng introduced the computation power development and the contradiction with AI demand. It was because of separation between Storage and Computation function.