Quality Alchemist (品質煉金術師): CAAI Chinese Congress on Artificial Intelligence 2019 (中国人工智能大会)

2019年9月21日星期六

CAAI Chinese Congress on Artificial Intelligence 2019 (中国人工智能大会) - Day 1

The 5th Chinese Congress on Artificial Intelligence (CCAI2019) which was led by Chinese Association for Artificial Intelligence (CAAI), was held on Sep 21-22, 2019, in Jiaozhou Qingdao, China. As the largest official AI conference in China, CCAI has been held annually to promote the advancement of artificial intelligence globally. The frame of congress named “Intellectual Change and Fusion (智变融合)”. The CAAI Extenics Professional Committee (中國人工智能學會可拓學專業委員會) representatives participated the CCAI 2019.

(Left: Mr. Jia Qingyu (贾庆玺), Dr. Lotto Lai, Prof. Xingsen Li (李興森), Mrs. Li and Mr. Zhang Zhiwang (張志旺))

We also took a group photo inside the venue.

So many experts attended the congress.

Before the congress, a video was played to introduce the CCAI and China AI policy.

Day 1 (21 Sep 2019): (Opening Session)
In the beginning, Mr. Xue Qingguo (薛庆国 – 中共青岛市常委，市政府党组副书记、副市长) chaired this session. Firstly, he introduced today’s guest speakers.

Mr. Wang Qingxian (王清宪 – 山东省委常委、青岛市委书记) gave guest speech. He said to hold CCAI as policy could be very successful to activate the city to modernize Qingdao. He called it as “AI Science and Technology Application Service Industry”.

Prof. Deyi Li (李德毅) (President of CAAI, Academician of CAE) gave a guest speech and he said manufacturing was the backbone of the national economic. AI could enhance smart manufacturing.

Prof. Justine Cassell (Associate Dean for Technology Strategy and Impact in the School of Computer Science at Carnegie Mellon University) gave a guest speech. She said Qingtao would incubate many startups on AI and the world focused the outcomes of AI.

Prof. Yan Kexiong (牟克雄 - 中科院自动化所党委书记) was the last guest to give speech and he said AI would be led a new growth and new round of the development of economy and society. We had set up the AI industrial technology research institution in Jiaozhou in 2018 and supported national strategy on AI.

The first speaker was Prof. Deyi Li (李德毅) (President of CAAI, Academician of CAE) and his topic entitled “The Future Transportation: Auto-Driving & Intelligent Networking (未来交通：自动驾驶与智能网联)”. Firstly, Prof. Li discussed how to implement auto-driving and he said the foundation of mass production included technology, cost, market and ecosystem.

The following diagram showed the estimation of auto-driving production from 1984 to 2045 from exploration stage to incubation stage and then to scaled development stage. However, Prof. Li said it needed application scenario such as auto-parking, point to point transportation, fast runway for public car and limited region for rental auto-driving taxi.

And then Prof. Li mentioned to establish auto-driving safety standard from level 1 to level 9. If it achieved level 4 both in auto-driving and safety, mass production would happen. Moreover, he added auto-driving needed to have learning capability and follow common sense.

After that he discussed the supply chain of auto-driving included auto-driving map, machine driver and control system. He also raised out the difficulties in the supply chain and many problems needed to be solved. Testing and management of auto-driving was important. Moreover, the legal responsibility should be concerned. For example, any accident of auto-driving happened, driver would take the responsibility if he in the driver seat of the auto-driving car; owner would take the responsibility if no driver in the auto-driving car.

Finally, Prof. Li discussed 5G improvement of internet ecosystem and enhancing auto-driving because of low latency and high connection power. Government needed to provide good ecosystem for future transportation. At the end, Prof. Li concluded AI could enable the future transportation and looking forward on 2030.

The second speaker was Prof. Zhang Wei (张钹 - 清华大学人工智能研究院院长中国科学院院士) and his topic named “A New Journey towards the Third Generation of Artificial Intelligence (迈向第三代人工智能的新征程)”. He said AI was not succeed through one battle, it was always on the road.

Firstly, Prof. Zhange introduced the first generation of AI that was Symbols theory based on knowledge-driven and searching. The benefit was explainable but only for limited area. The second generation of AI was Deep Learning and its breakthrough was no need of domain knowledge but based on big data. However, Prof. Zhang though AI and Big Data in AI era was not equal to Stream Engine and Computer as previous industry revolution because of limited application scenario.

Then he compared human brain and AI system based on deep learning for energy consumption, robust, extension capability, big mistake percentage, training data sample and explainable. AI system based on deep learning was still weak. If we added some noise together with photo, AI system would identify wrong object but it never happened in human.

Lastly, Prof. Zhang introduced the third generation of AI that should be integrated with Data-driven and Knowledge-driven system. That included deep learning and knowledge graph fusion in natural language processing, small sample learning, cause and effect reasoning, etc.

The third speaker was Prof. Justine Cassell (Associate Dean for Technology Strategy and Impact in the School of Computer Science at Carnegie Mellon University) and her presentation title named “Why Social AI?” In the beginning, she said society should be collaboration between humans and machines and not the replacement of human by machines in the future.

Then she introduced virtual peer learning approach to help children learned more science and to feel rapport with this virtual peer. Rapport means that a close and harmonious relationship in which the people or groups understand each other’s feelings or ideas and communicate well.

After that she discussed why socially aware AI and Chatbots aimed to build rapport but failed because people changed interaction styles over time. Then she introduced the methodology to build a study model and demonstrated some students didn’t like mathematics, but the virtual peer could make fun on mathematics and good conversation with students.

At the end, Prof. Justine Cassell demonstrated their system and students’ response to us. Student behaved funny with virtual peer. She concluded that Social science, AI and Human-Computer Interface had overlapped area to form a Human-Agent Interface.

The forth speaker was Prof. Zhang Qin (張勤 - 全国政协常委，中国科协荣誉委员) and his presentation topic was “Smart Medical promote the Grading Diagnosis and Treatment – Preliminary Practice of DUCG (Dynamic Uncertain Causality Graph) (智慧醫療促進分級診療 – DUCG的初步實踐)”. Prof. Zhang said we was not able to make a right decision on diagnosis because of information changing dynamically.

Then Prof. Zhang said there were two directions to research that one was big data learning and the other was traditional system. Basically, he established 11 knowledge databases that certified by the third party. Another 20 knowledge databases were under development and testing.

Using this DUCG could enhance the medical and diagnosis level in rural area in China.

After that he demonstrated the DUCG system and e it could also employ in Chinese Traditional Medical diagnosis.

Finally, Prof. Zhang concluded that DUCG no needed special doctor feedback from each case but learn from the huge past diagnosis results for reversed calculation so as to enable general doctor in rural area and enhanced the level of diagnosis. He also demonstrated the structure of patient-based cloud DUCG ecosystem at the end.

The last speaker in the morning session was Dr. Xiao Jing (肖京 - 中国平安集团首席科学家) and his presentation topic entitled “Intelligence + Financial Strategy Practice (智能+金融战略实践)”. Firstly, Dr. Xiao introduced their “Finance and Escosystem” Strategies included new technology such as ABCDS (i.e. AI, Blockchain, Cloud, Big Data and Security).

Then he stated the existing pain points in the Financial Industry on customer, service, operation and risk control. He mentioned from “Internet +” to “Intelligence +” establishment. Some industrial scenarios were focused such as smart finance, smart medical and smart city.

After that Dr. Xiao briefed the Federal Transfer Learning could handle data isolated island and protected security of data.

Lastly, Dr. Xiao briefed the AI Ethic structure in his company included Self-control, Safety, Fair and Open.

After the morning session, we went to the lunch hall and a interested couplet (對聯) mentioned “天地人生智變融合”、“人生非正非反正正反反反反正正反正一生” ; “天地孰方孰圓方方圓圓圓圓方方方圓九州”.

We enjoyed a nice rice box.

In afternoon, there were two parallel session and I joined the session named “AI Young Forum: Do Deep Learning meet the bottleneck?” (人工智能青年论坛：深度学习到瓶颈嗎？). Prof. Yu Yang (俞扬 - 南京大学人工智能学院教授国家"万人计划"青年拔尖人才) chaired this session.

The first speaker in this session was Dr. Jingdong Wang (王井东) (IAPR Fellow, ACM Distinguished Member, Senior Principal Research Manager, Microsoft Research, Beijing, China) and his presentation entitled “Deep High-Resolution Representation Learning for Visual Recognition”. He said convolutional neural networks were good at representation learning.

Dr. Wang explained previous classification networks were connected multi-resolution convolutions in series but High-resolution networks (HRNet) were connected multi-resolution convolutions in parallel with repeated fusions.

And then he showed visual recognition applications included image classification, object detection, semantic segmentation, face alignment and pose estimation. Finally, he summarized to maintain high-resolution representations through the whole process with repeated across-resolution fusions.

The second speaker was Prof. Yan Jingyi (虞晶怡 - 上海科技大学信息科学与技术学院执行院长) and his topic named “Learning to Build a New Reality”. He introduced traditional computer vision that lack of stereo parallax. Then he employed passive 3D imaging and early attempted was VR for medical training.

Prof. Yan introduced “Sensing + Cognition” (感知 + 認知) using a learned prior that can use a photo reformed 3D image.

Finally, he briefed the image reformation from face to body that learn to reconstruct 3D Sequence. Lastly, Prof. Yan summarized that “Sensing + Cognition” was equal to the “New Normal”. Learning-based techniques would produce superior 3D.

The third speaker was Prof. Meng Deyu (孟德宇 - 西安交通大学教授国家"万人计划"青年拔尖人才) and his presentation entitled “Model-driven vs Data-driven in Underlying Vision Task” (底层视觉任务中的模型驱动对数据驱动).

Firstly, Prof. Meng discussed the advantages and disadvantages between Model-driven and Data-driven for vision task. Model-driven used small sample and model focused but low prediction speed and difficult design. Data-driven was no need to test hypothesis or design model but poor in explanation and based on labelled data, as well as, difficult in structure design.

Then Prof. Meng proposed to integrate both driven approach together as follows:

i) Semi-supervised Deep Learning

ii) Deep unfolding between algorithm and network

iii) Using model-driven methodology found generative understanding and then network parameters were shared by posteriors calculated on all training data so as to minimize divergence through variational inference.

The forth speaker was Prof. Zhang Zhaoxiang (张兆翔 - 中科院自动化所研究员国家"万人计划"青年拔尖人才) and his presentation topic named “Deep Learning Based Object Detection: Our Recent Progress”. Prof. Zhang reviewed object detection before Deep Learning using sliding windows and deformable part models (DPM).

Deep learning greatly benefits the visual object detection and it still had potentials to develop new models and methods. Prof. Zhang briefed three types of approach and they were “Scale-Sensitive Practical Object Detection (POD)”, “Efficient Neural Architecture Search for Object Detection (NATS)” and “Trident Network for Object Detection (TriNet)”.

The fifth speaker was Dr. Su Hang (苏航 - 清华大学助理研究员) and his topic was “Deep Learning against Security Theory and Method” (深度学习对抗安全理论与方法). He said to ensure the safety and security of AI system was focused by different countries in the world.

Dr. Su used face recognition as example that people found the model characteristic so as to attack it such as adding noise. In order to prevent such attack, he proposed two strategies. The first one was to filter the noise and reduced noise interference and the second one was to design new model structure and loss function so as to enhance robust.

Finally, he proposed a TRUE AI, indicating that Trustworthy, Robust, Understandable and Ethical. That is the third generation of AI.

The sixth speaker was Prof. Qiu Xipeng (邱锡鹏–复旦大学副教授) and his presentation entitled “Learning Progress from Transformer to BERT – Natural Language Processing” (从Transformer到BERT - 自然语言处理的表示学习进展)

Prof. Qiu mentioned three difficulties about Natural Language learning included Model level, Learning Level and Cognition Level. Then he introduced the combined model for NLP involved CNN, RNN and Transformer.

After that we went to another parallel session named “Smart City Forum” (智慧城市论坛).

Mr. Jia Qingyu (贾庆玺), Prof. Xingsen Li (李興森) and I took a photo in front of the venue.

The first speaker in the Smart City Session was Prof. Zhang Wensheng (张文生 - 中国科学院大学人工智能学院首席教授) and his topic entitled “AI in the era of Big Data and its Application” (大数据时代的AI及其应用). Firstly, he shared four views that were “How to quantify intelligence?” “How to understand AI?” “Use feedback from real intelligence” and “Big Data enhance intelligence”.

Prof. Zhang reviewed the new generation AI development policy and other research report and pointed out four areas such as Finance, Vehicle, Retail and Medical that were affected by AI significantly. Moreover, Manufacturing, Education and Telecommunication industries were also concerned.

The second speaker was Prof. Chen Nengcheng (陈能成 - 武汉大学教授) and his topic named “Urban Multi-scale Integrated Perception and Service” (城市多尺度综合感知与服务). Prof. Chen introduced his approach named CityI³Sensing that included Integrated, Intelligent and Instant.

The following diagram summarized different integration applications in different areas.

The third speaker was Prof. Xue Guangtao (薛广涛 - 上海交通大学信息学院院长) and his presentation was “Research on Key Technologies of Heterogeneous Service Synergy” (异构服务协同共性关键技术研究). Prof. Xue mentioned the city development in China was speeding and upto 5 million population city increased continuously. Therefore, city service demand were under pressure.

There were some key problem observed and they were not connected in system, data and service.

Finally, Prof. Xue introduced his key focused research items from model research to technology breakthrough and then for scale demonstration.

The forth speaker was Prof. Cheng Xiuzhen (成秀珍 - 山东大学计算机学院院长 IEEE Fellow) and her presentation named “The Challenge of High Confidence Smart City Service” (高置信智慧城市服务挑战). She said high confidence smart city service was very huge that bigger than smart houseware and transportation. It related to national and social security.

Prof. Cheng also stated to use blockchain for high confidence smart city service so as to enhance security and reliability.

Lastly, she raised some challenges in Internet, Big Data and Machine Learning, as well as, Blockchain. She expected to form standardized structure to cover all aspects in the high confidence smart city service.

The fifth speaker was Dr. Feng Dahang (冯大航 - 声智科技联合创始人兼CTO) and his topic was “Progress and Application of Acoustic and AI Fusion Technology in Smart Cities” (智慧城市中声学与AI融合技术的进展与应用)

In the beginning, Dr. Feng reviewed the smart city development in China and found that the investment increased annually. Then he pointed out two trends that were “Acoustic + AI fusion” and “New structure framework establishment named Octopus (章魚) system”.

Lastly, Dr. Feng briefed their Octopus system that was multi-sensor driven new network to service transportation, security, houseware and city administration.

The last speaker in this session was Dr. Shan Zhiguang (单志广 - 国家信息中心信息化和产业发展部主任) and his presentation title named “Smart Transformation of Smart City” (智慧城市的智慧转型). He said there were 1000 smart cities in the world and 500 of them were in China.

The following diagram showed the industries distribution under China smart city. Then Dr. Shan briefed the content of smart city included system capability, characteristic, structure, economic status, carrier feature, design principle, network service model, technology implementation roadmap, and operation model.

Lastly, Dr. Shan introduced the Evaluation of National New Smart City. He used word “3D” (Digit Determines Destination -数字决定命运) as conclusion.

Finished the CCAI 2019 day 1 event, I was honor to be invited by Prof. Xingsen Li (李興森) to join his dinner with his good friends, classmates and colleagues in Qingtao.