Quality Alchemist (品質煉金術師): CAAI Chinese Congress on Artificial Intelligence 2018 (中国人工智能大会)

2018年7月28日星期六

CAAI Chinese Congress on Artificial Intelligence 2018 (中国人工智能大会) - Day 1

The 4th Chinese Congress on Artificial Intelligence (CCAI2018) which was led by Chinese Association for Artificial Intelligence (CAAI), was held on July 28-29, 2018, in Shenzhen, China. As the largest official AI conference in China, CCAI has been held annually to promote the advancement of artificial intelligence globally. I attended the Congress and summarized it for sharing AI trend to all quality professionals.

During the congress, I met the Prof. Deyi Li (李德毅) (President of CAAI, Academician of CAE) and took a photo for memory.

I also met Prof. Qiang Yang (杨强) (Vice President of CAAI, AAAI/ACM/IEEE Fellow, President of IJCAI; Chair Professor, Department Head of CSE, HKUST).

Day 1 (28 July 2018):

In the beginning of the congress, Prof. Deyi Li (李德毅) (President of CAAI, Academician of CAE) gave an opening speech and shared the AI Trend in China.

Mr. He Haitao (贺海涛) (深圳市人大常委会党组成员、罗湖区委书记) gave guest speech and supported AI industry in Shenzhen

Mr. Wang Lixin (王立新) (广东省深圳市副市长) gave guest speech and he said his mind though Artificial Intelligence after he was on board in Shenzhen.

Prof. Tie-niu Tan (谭铁牛) (Vice President of CAAI; Academician of CAS) gave guest speech and he said using four Chinese words for coming AI development that was “理性务实” (meaning Rational and Pragmatic).

Day 1 Congress reporting session named “The Future of AI is at Hand”, Prof. Chenqing Zone was session chair and introduced the first speaker.

The first speaker was Mr. Harry Shum (沈向洋) (Executive Vice President of Microsoft’s AI and Research group; ACM/IEEE Fellow) and his topic named “The Future of AI is at Hand and Thinking about Xiaobing (小冰) product implementation”. He said that AI developed very fast recently because of big computing, big data and precise model.

Then Mr. Harry Shum briefed Microsoft Research AI breakthroughs included 96% RESNET vision test 152 layers in 2016, 5.1% Switchboard speech recognition test in 2017, 88.5% SQuAD reading comprehension test and 69.9% MT research system in January and March 2018, respectively. And then he briefed Microsoft Xiaobing (小冰) natures included dialogue platform, Bot Factory, EQ+IQ, AI Creation and Business Opportunity.

Mr. Harry Shum mentioned the Emotion Calculation Framework included human perception, Dual AI and World context.

Microsoft Xiaobing (小冰) focused to break through the Emotion Calculation and established relationship with users. It is one of largest user based AI dialogue system in the world with 660M users (120M active users per month). The evolution of line would be from EQ+IQ to AI Creation. (From Retrieval Model → Generation Model → Empathy Model)

After that Mr. Shum stated the Three Principles of AI Creation as follows:

i) Integrated EQ and IQ (Not only IQ);

ii) Product created by AI should have independent IP right;

iii) The process of AI Creation should correspond to human creativity, not only simplify human labour work.

Mr. Shum demonstrated AI sang a song.

At the end, Mr. Harry Shum mentioned the concept “Intelligent Cloud” that means calculation anywhere and Intelligent anywhere – The world is becoming a computer. Lastly, he quoted Bill Gates that “Most people overestimate what they can do in one year but underestimate what they can do in ten years.”

The second speaker was Prof. Qiang Yang (杨强) (Vice President of CAAI, AAAI/ACM/IEEE Fellow, President of IJCAI; Chair Professor, Department Head of CSE, HKUST) and his topic named “The Challenges that AI Faces and the Opportunities that Transfer Learning Brings (Federal Transfer Learning)”. He discussed Big Data driven AI ideal and fact, as well as, AI history in raising and fall.

Then Prof. Yang told about data from different companies had different characteristics that was difficult to communicate each other. AI’s Big Data had two difficulty situations that they were privates (especially EU General Data Protection Regulation - GDPR) and small data. He quoted Prof. Pedro Domingos (29 Jan 2018) that Starting May 25, the European Union will require algorithms to explain their output, making deep learning illegal!

And then Prof. Yang introduced Federated Transfer Learning (FTL) (聯邦遷移學習) to establish machine learning (ML) of enterprise ecosystem. It could solve the data security problem through FTL framework.

“FLT had Heterogeneous Transfer Learning to find latent representations and Homomorphism-aware loss function via Polynomial Approximation, as well as, Privacy-Preserving Entry-id Match.” Prof. Yang said. Finally, he said establishment of security big data plus AI ecosystem such as Financial Federated Learning Network that solved GDPR privacy requirement.

The third speaker was Mr Jason Dai (戴金权) (Senior Principal Engineer and CTO of Big Data Technologies at Intel) and his topic named “An Introduction to Analytics Zoo for Apache Spark and BigDL”.

He would like to introduce the trends of AI as follows:

i) Data Scale Driving Deep Learning Process

ii) Hadoop Becoming the Center of Data Gravity

iii) Real-World ML/DL Systems are Complex Big Data Analytics Pipelines

iv) Unified Big Data Platform Driving Analytics & Data Science

Then he introduced Unified Big Data Analytics Platform which consisted Apache Hadoop and Spark Platform. Mr. Dai also mentioned challenges of productionizing Large-Scale Deep Learning solutions as follows:

i) Very complex and error-prone in managing large-scale distributed systems

ii) Low end-to-end performance in GPU solutions

iii) Very inefficient to develop the end-to-end processing pipeline

Lastly, he was talking about neural recommendation engine in China Life and summarized to make deep learning more accessible to big data and data science communities and using Analytics Zoo for end-to-end analytics plus AI platform for Apach Spark and BigDL.

The forth speaker was Mr. Zhiqiang He (贺志强) (Senior Vice President of Lenovo Capital and Incubator Group) and his presentation topic was “Multiple-speed Industry Opportunities in Smart Internet Era”. He said Lenovo Investment aimed to effective capital and entrepreneur’s power that included VC and Incubation.

He then briefed the path from AI to Intelligent Internet including Internet + boundary calculation + cloud + big data + AI that integrated into industry. He also said high quality data was the base of AI.

After that Mr. He said core of AI was Data and Algorithm which immersed in cloud, terminal and chip. He also said to seek the investment opportunities on intelligent internet core technology as following diagram.

At the end, Mr. He said to plan for Great Bay Area and establishing Lenovo Open Platform for value chain acceleration service.

In afternoon session, Mr. Tong Zhang (张潼) (Director of Tencent AI Lab) was session chair.

The fifth speaker was Prof. Liwei Wang (王立威) (Professor of Peking University) and his presentation topic entitled “Machine Learning and Artificial Intelligent”. He would discuss the core technology of Machine Learning (ML) and supervised learning, as well as, trend of ML development.

Firstly, Prof. Wang said ML is not a new things and he review some physics law such as Hooke's law and Kepler's laws which used for prediction. Therefore the core framework of ML is to collect data for learning model and then predict the new data.

Then Prof. Wang stated the nature science used simple equation to describe the world but ML designed complex model which learn from Big Data to solve complicated problem. He also briefed the Supervised Learning in which Generalization Capability (泛化能力) is the core concept. Their goal is to make generalization error as small as possible.

After that Prof. Wang briefed the history of ML from 1943 to 2006. Neural Networks is revitalized through Deep Learning from 2006 to now, because of layer by layer training methodology, hardware computing power increase significantly and big data. AI success implementation areas included graphic, voice, translation and game. However, AI only suitable implement in the close loop knowledge with narrow scope.

Finally, Prof. Wang discussed the trend of AI was not limited to solve narrow scope problem. If no breakthrough of AI development, we were not able to solve problem in open environment employing common sense. Now, we needed to train up industry habit for data collection and standardization especially in medical health industry and financial industry.

Forum II – Smart Chip: Industry Promotion and Innovation Development

Prof. Yinhe Han (韩银和) (Professor of ICT, Chinese Academy of Sciences (CAS)) was session chair and he firstly shared the topic Smart Chip that chip could provide special intelligent capability focused on deep learning.

Prof. Han mentioned the smart chip status from inference on device, on cloud and for training. He also said CAS proposed “DianNao” neural network processor unit in Feb 2014 and then their spin off company was acquired by Xinlinx (FPGA factory) in Jul 2018.

The first guest of forum was Prof. Yunji Chen (陈云霁) (Professor of ICT, CAS) and he said that education was most important to promote smart chip. He explained Smart Computer that could provide CPU and intelligent accelerator for developers’ programming.

And then Prof. Chen introduced smart computer course and it aimed to achieve “Application Drive and Full Stack Through (應用驅動、全棧貫通)”.

The second guest was Dr. Yu Wang (汪玉) (Associate Professor of Tsinghua University; Cofounder of DeePhi) and his sharing topic named “Neural Networks on Chip: From CMOS Accelerators to In-Memory-Computing”. Firstly, he said AI is a large area for business scope and introduced Software and Hardware co-design which obtained the best performance.

After that he mentioned Neural Network Accelerator Inference, Deep Learning Platform to Vertical Markets and DeePhi Compression Tool, as well as DeePhi Reference Algorithms.

Lastly he demonstrated their research and application. He also discussed the energy efficient should be considered for smart chip.

The third guest was Dr. Jiaen Liang (Founder of Unisound) and he shared title was “Internet of Things “Chip Era” – Open Source Solution from Module to Chip”. He introduced AIoT chip named UniOne. Their SoC included CPU (ARM), AI accelerator (DeepNet) and Digital signal processor (DSP).

Then he introduced their product named UniOne smart speaker and family solution.

The fourth guest was Prof. Xin Li (李昕) (Professor of Duke University; IEEE Fellow) and his topic named “The missing Components in AI Chips”. In the beginning, he introduced AI supply chain.

Then he mentioned 3 challenges we met below:

i) Efficient Programming for Hardware Accelerators

ii) Efficient Validation for Integrated Products

iii) Efficient Adoption for Broader Applications (Business translators are important!)

Finally, he showed a 30-year R&D plan from 2020 to 2050.

The fifth guest was Mr. Wenyuan Dai (戴文渊) (Founder of 4Paradigm, CEO) and he explained the company name which was come from Project Jim Gray’s scientific development from first paradigm to fourth paradigm which depended on data science.

He said the integration of hardware and software from two stages. First stage is training included tasks of graphic and decision; and the second stage is predication included tasks of cloud and terminal. He concluded that AI algorithm and application should be linked with server and chip.

The last guest was Mr. Zhiqiang He (贺志强) (Senior Vice President of Lenovo Capital and Incubator Group) and he said Lenovo Capital invested in smart internet since 2016.