Reports from the Heart of the Machine
Machine Heart Editorial Department
When in the middle of spring, Yang and Fangqi. On March 23, the Heart of machine AI Technology Annual Conference was successfully held in the form of online live broadcast.
At this event, we set up three forums: the Artificial Intelligence Forum, the AI x Science Forum and the Chief Intellectual Officer Conference, and invited 30 heavyweight guests to fully communicate around a number of the most valuable topics at present.
Although they could not meet offline, everyone was enthusiastic: on the day of the event, the live broadcast was watched by more than 26,000 people. The wonderful discussions and views of the guests have also aroused heated discussion among online audiences.
12 guests talked about the Artificial Intelligence Forum
As the first guest to appear at the Artificial Intelligence Forum, academician Zheng Weimin shared the theme of "FABS: Computing Models integrating artificial intelligence, big data and scientific computing".
In recent years, intelligent computing is accelerating its integration with traditional scientific computing, and remarkable progress has been made in protein structure prediction, weather forecasting, and molecular dynamics. Both AI and scientific computing rely on data processing, but existing Intelligent + Scientific Computing (AI-HPC) systems mainly use the programming mode of MPI+X to express the complexity of data processing tasks, while adding a data processing system such as Spark or Pandas faces challenges in terms of system complexity, performance or cost. In addition, MPI+X's fault tolerance is relatively poor, relying on global checkpoints and recalculation techniques, and when the system scale is extended to E-level and post-E- level, the average failure-free time of the whole machine is only a few hours, which poses a major challenge to the effective use of the machine.
As a result, there is always a lack of programming that effectively expresses HPC+AI+BigData. Based on this, Academician Zheng Weimin proposed a computing model that integrates artificial intelligence, scientific computing and big data processing (FABS: Fused AI, Big Data and Science), through unified tensor abstraction and compilation optimization technology, while providing easy-to-program, highly available, high-performance programming models and computing models for these three fields, which will provide important tools for the development of large-scale AI+Science.
Next, Deng Li, former chief scientist of Microsoft AI and chief artificial intelligence officer of the Castle Fund, shared his practical experience in speech language, financial investment, online education and health care. In recent years, artificial intelligence technologies, including deep learning, have completely subverted the global speech recognition and natural language processing industry, and have also brought great impacts to the financial investment industry, and have achieved initial outstanding results. In addition, in other fields such as online education and medical health, natural language processing technology based on deep learning is also becoming a mainstream approach.
In order to achieve wider success and adoption, there are several technical challenges that need to be addressed, such as pre-training and self-training of models, how to do transfer learning, etc., which will be very helpful for progress in areas related to low samples and noisy labeling data, such as the medical and financial industries. Deng Li pointed out that another big challenge comes from adversarial learning, that is, adversarial deep learning for multiple agents, taking stock prediction as an example, in the statistical distribution of the stock market, today and tomorrow may be very different. In order to solve the problem of adversarial competition, more progress is needed in this area of technology.
Yiran Chen, a professor in the Department of Computer Engineering at Duke University, shared the theme of "Software and Hardware Collaborative Design of Efficient Artificial Intelligence Systems". Over the past 100 years, computing power has shown an exponential growth trend, and the possibilities are endless. There are many kinds of computing platforms for AI, but whether it's GPUs, FPGAs, ASICs, or other new architectures, they basically follow the same principle: more efficient, or take longer; more professional, or more flexible, it is actually difficult to achieve unity across multiple dimensions.
In the face of such contradictions, Over the years, Professor Chen Yiran's team has done a lot of relevant research accumulation, starting from 2012 to studying the expression of different hardware, and later doing architecture design, distributed design, and even automated design. At the same time, Professor Chen Yiran also pointed out that there are still many opportunities and challenges in the design of full-stack efficient artificial intelligence systems, and many directions of work need further research.
Subsequently, Zhou Jun, general manager of the financial machine intelligence department of Ant Group, shared the theme of "The Practice and Exploration of Trusted AI in the Digital Economy".
Zhou Jun introduced that if the digital economy is compared to a tree, artificial intelligence (AI), big data, cloud computing and other technologies in the trunk constitute the core of the digital economy and play a role in carrying forward the upper and lower levels; factors such as privacy and security in the roots of the tree determine the growth trend and the future; the trunk and the root must be closely integrated in order to flourish, of which AI + privacy, AI + security, etc. have become the direction of urgent breakthrough. The concept of trusted AI technology will be one of the key capabilities to resist risks and enhance the inclusiveness of science and technology in the digital age, and the system has made many research breakthroughs and landings in the technical directions of privacy protection, explainability, and confrontation, and still has a long way to go and needs continuous investment.
In the direction of graph machine learning, Ant Group proposed the graph learning system AGL, which can support the graph data structure of industrial scale and help identify transaction risks; in the direction of fairness, it proposes SMEs (small and medium-sized enterprises) credit score, through graph learning, fusing multi-source information, mining potential complex patterns, and helping SMEs enjoy financial services; in the direction of interpretability, a model-independent interpretable method COCO is proposed, and counterexample samples are obtained through information weighting for limited disturbance, and then by measuring counterexample samples. Calculate the characteristic importance of test samples to give the interpretability of arbitrary models; in addition, Zhou Jun shared that Ant Group combines distributed machine learning with dense state computing and proposes a privacy protection machine learning method CAESAR, which can further improve computing efficiency and reduce traffic in the case of enhancing information protection strength. Zhou Jun concluded that the trusted AI technology concept built around privacy protection, robustness, explainability and fairness will continue to promote the transparency and friendliness of artificial intelligence technology in the digital economy scenario, make decision-making more intelligent, and make the digital economy deeply intelligent.
Subsequently, Zhang Faen, CTO and co-founder of Innovation Qizhi, introduced the work of Innovative Qizhi in the commercialization of artificial intelligence technology, including vision-related and structured machine learning technologies, and the MMOC (MenuVision, MatrixVision, Orion, Cloud) platform created by Innovative Qizhi.
In the final speech of the morning, Xie Yutao, engineering director of IDEA Research Institute and head of the AI Platform Technology Research Center, shared his thoughts on academic research tools and new research ecology. Under the wave of new technologies, each node in the scientific research ecology has a lot of room for optimization and iteration. Xie Yutao took the paper community Readpaper as an example to show the audience of this forum an efficient and professional new academic community. Among them, the functions of paper search, literature management, reading and academic exchange groups have become scientific research tools for early users.
In the final part, The Machine Heart Pro provides a brief look at some of the content of the upcoming global AI Technology Trends Report, which will be released in April. At present, the project team of "2021-2022 Global AI Technology Trend Development Report" has basically completed the basic work of academic literature included in 11 international top conferences, hundreds of well-known digital transformation and technological innovation projects in recent years, and data analysis of nearly 100 AI development tools, and completed most of the basic research work combined with targeted expert interviews.
In the afternoon forum sharing, Yang Qiang, chairman of the FATE Federal Learning Open Source Community Technical Committee, first brought the theme sharing of "Trusted Federated Learning", systematically reviewed the progress and challenges of federal learning, and looked forward to several important development directions.
Yang Qiang pointed out that today's AI still has a bottleneck of over-reliance on centralized data. In the real world, data often shows the characteristics of multi-source, dispersion, and large changes, and the development and application of privacy computing technology is becoming more and more interesting. Among them, trusted federated learning has the characteristics of safe and provable, performance usable, controllable efficiency, interpretable decision-making, supervisable model, and inclusive. In the past two years, federated learning has been included in gartner's technology maturity curve (one of the most valuable reports on new trends in technology in the world), and federated learning in the embryonic stage of technological innovation is receiving increasing attention as the key to the development of the next generation of private computing.
Fate, the world's first privacy computing and federated learning open source community, was born, and at present, FATE has attracted 3,000+ engineers and developers, 800+ enterprises, 350+ universities, and 3,200 GitHub Stars. According to the survey statistics of the Chinese Academy of Information and Communications Technology, 55% of domestic privacy computing products are based on or reference open source projects, of which the FATE open source project is the mainstay.
Zhou Ming, chief scientist of innovation workshop, founder of Lanzhou Technology, vice chairman of the Chinese Computer Society, and former president of the International Computational Language Society, shared the theme of "The Innovative Era of Cognitive Intelligence". He introduced Lanzhou Technology's next-generation cognitive services engine program, including lightweight pre-training models and progress in natural language understanding and generation, and shared views on its future development trends and commercial landing.
At present, AI is rapidly moving from perceptual intelligence to cognitive intelligence. AI is moving from being able to speak and see, to being able to think and answer questions, to making decisions and reasoning. Facing the development trend and industrial background of cognitive intelligence, Lanzhou Technology, which is incubated by the innovation workshop, proposes The research plan of Mencius's new generation of cognitive service engine, with the goal of studying the core task of cognitive intelligence and promoting the digital transformation of the industry with cognitive intelligence technology. Lanzhou Technology has developed Mencius's lightweight pre-training model, as well as advanced machine translation, text generation and industry search engines built on it, and empowered industry customers through open source, SaaS and customization.
Zhou Ming pointed out that in the next decade, AI will leap from perceptual intelligence to cognitive intelligence, benefiting human society. The future development direction, on the one hand, is to solve the core technology of model distillation, compression and lightweight model along the extension line of pre-training, and reduce the problems caused by data deviation and privacy; on the other hand, it is also necessary to solve some important problems in the algorithm, including the integration of neural network systems and knowledge systems, the study of better small sample learning mechanisms, the activation and establishment of common sense, and the interpretable mechanism.
Jingyi Yu, Vice Provost of ShanghaiTech University and Professor and Executive Dean of the School of Information Science and Technology, brought wonderful content sharing about digital people. In recent years, digital people have become one of the hottest technology concepts. Yu Jingyi said that there are currently several trends in this field. First of all, the traditional 3D reconstruction classical algorithm is gradually being replaced by deep learning-based algorithms. At the same time, good reconstructions have been replaced by good renderings. The original reconstruction must have a good three-dimensional geometry. Now, the images generated by the neural network rendering are enough to achieve similar or even better visual effects. Most importantly, the entire field is evolving from explicit expression to implicit expression. The original 3D reconstruction was about point clouds, maps, BRDF lighting, and now it's about NeRF, NeuS, and NPG. In the future, replacing the traditional explicit expression with an implicit expression similar to neural networks is expected to become the focus of 3D vision and its research in the direction of virtual reality and metaversity.
Yu Yang, professor at the School of Artificial Intelligence of Nanjing University and founder of Nanqi Xian Ce, shared the relevant content of the theme of "Bringing Reinforcement Learning Superman's Decision-making Ability into Reality". At present, reinforcement learning technology has achieved universal decision-making ability beyond humans in tasks such as Go and games, and we are very much looking forward to reinforcement learning also landing in practical applications, so that we have strong decision-making ability. One of the obstacles to achieving this goal is that existing reinforcement learning techniques lack the imagination of humans in general, and can only find optimal decisions from a large number of trial and error, and the game just happens to provide a large number of trial and error possibilities. In his speech, Professor Yu Yang introduced his work in the direction of making reinforcement learning imaginative, and the application of reinforcement learning in real business.
Subsequently, Wei Yichen, vice president of R&D of Shukun Technology, brought a keynote speech on "The Application and Exploration of AI in Medical Imaging". In recent years, the application of AI in medical imaging has gradually matured and landed, and the industry is changing rapidly. Due to the particularity of the medical industry, the threshold of AI products is high, and the research and development process is also more complicated. In the speech, Wei Yichen introduced the characteristics, status quo and future of product research and development in the industry.
Graph neural networks and geometric deep learning are emerging directions of deep learning, which have important applications in the fields of protein prediction, new drug design, mathematical theorem proof and discovery, and are important models for trusted artificial intelligence. At the end of the forum, Wang Yuguang, associate professor of the Institute of Natural Sciences and the School of Mathematical Sciences of Shanghai Jiao Tong University, introduced the research progress and development trend of geometric deep learning and graph neural networks.
What was talked about at the CJI Conference?
Today, Auto Byte, a subsidiary of Machine Heart, has been established for more than a year, and as an information platform focusing on smart mobility, it has also held a parallel forum - the "Chief Intellectual Officer Conference". At this conference, Auto Byte invited a number of leaders from OEMs, autonomous driving companies, and chip companies to set up five theme sharing and two roundtable forums, which received widespread attention from inside and outside the industry.
The guests of the conference include: Xia Yiping, CEO of Jidu Automobile, Gu Weihao, co-founder and CEO of Zhixing, Wang Ping, EXECUTIVE PRESIDENT of Cambrian Xingge, Xiao Jianxiong, founder and CEO of AutoX, Li Bo, vice president of Lotus Technology and head of intelligent driving business line, Yang Yuxin, chief marketing officer of Black Sesame Intelligence, Wang Kai, director and CEO of Xinqing Technology, Zhou Xin, co-founder and chief product officer of Yishi Technology, Hao Jianan, co-founder and chief architect of Tucson Future, Dong Jian, co-founder of Hongjing Intelligent Driving, and Dai Zhen, vice president of Heduo Technology, attended the event with a total of 11 guests. Everyone conducted in-depth analysis and exchanges on the various fields of smart travel that are currently hot.
In the speech session, Xia Yiping, CEO of Jidu Automobile, shared an internal research data: the time spent by current users in the stationary state of electric vehicles has been equal to or exceeds the time spent driving. Electrification + intelligence is making the car a second living space, AI has brought technological innovation, efficiency improvement and experience subversion, 2023 will be the first year of automotive intelligent competition.
He also mentioned that the era of intelligent car 3.0 has arrived, and the characteristics of free movement, natural communication and self-growth of Jidu automobile robots are the product characteristics of this era. In addition, the intelligent car 3.0 era will also pay more attention to software security, Jidu Automobile has also developed its own electrical and electronic architecture and domain controller to ensure overall security from the perspective of combining software and hardware.
In this field, the artificial intelligence technology company dedicated to autonomous driving also has rich experience. Gu Weihao, co-founder and CEO of the company, said that data intelligence is the most fundamental driving force for the evolution of autonomous driving AI, and better algorithms and service models can be trained through further learning and mining of feedback data OTA to the car side, which can bring better system performance to users. In this process, cost and speed are the two most critical aspects, and they are also the ideological seal of data intelligence.
Gu Weihao believes that in the autonomous driving industry, who can efficiently and cost-effectively mine the value of data can become the king of competition. Data intelligence is the core of the evolution of AI autonomous driving technology, and a perfect data intelligence system is the cornerstone of the success of AI autonomous driving technology companies. Up to now, the mileage of assisted driving users has exceeded 6 million kilometers.
When it comes to autonomous driving, the hottest hardware topic at the moment is chips. As a chip company focusing on the field of autonomous driving, Cambrian Xingge's planning route has attracted much attention. Wang Ping, CEO of the company, also talked about the multiple challenges faced by the large-scale landing of intelligent driving on the chip in his speech: for example, the current problem of insufficient single-chip processing capacity has led to a significant increase in the complexity of the system. The multi-chip solution will cause the power consumption of the pre-controller to be high, and it must be air-cooled or even liquid-cooled, which increases the system cost and is difficult to popularize in fuel vehicles or economic electric vehicles. In addition, he also mentioned that the proportion of domestic chips is still low, and the overall supply chain is still facing greater challenges.
When it comes to the future trend of autonomous driving chips, Wang Ping also gave two judgments: one is universal openness, and the other is large computing power. He said that in the era of L1 and L2 automatic driving, because the amount of data is relatively small, many car companies can accept the closed integration scheme of strong coupling of chips and algorithms, but the amount of data in the L3 and L4 eras has surged, and the algorithm is more complex, and it needs large computing power chips to meet the demand. In the future, Cambrian Will also launch products covering different levels of autonomous driving, including SD5223 (this year) and SD5226 (next year) that will be launched this year and next year. Among them, SD5223 is a product for the L2 + market, the maximum computing power exceeds 16 TOPS, and a single SOC can realize the function of the integration of row and berth; SD5226 is for the L4 market, supporting the product of car-side training, using a 7nm process, AI computing power of more than 400 TOPS, CPU maximum computing power of more than 300K+DMIPs.
As an ultra-luxury brand that is undergoing a new transformation, Lotus Technology also shared the industry and technical ideas: Li Bo, vice president of Lotus Technology and head of the intelligent driving business line, said that in Lotus's internal user interviews, intelligent experience has surpassed configuration, brand and service, becoming the primary factor affecting the purchase of high-end luxury cars of more than 700,000 yuan, and among them, intelligence accounts for a large proportion of redefining luxury cars. In the era of intelligence, the ability of intelligent driving system has begun to replace power performance and become the most critical part of pure electric intelligent vehicles.
Li Bo also said that Lotus will redefine the intelligent driving classification with takeover mileage and coverage mileage, with the goal of creating end-to-end intelligent driving covering high-speed expressways, urban roads and parking scenarios, and at the same time creating Lotus's unique "track-level intelligent driving" with higher precision full coverage perception ability, better game-aware cognitive ability, and faster and more stable control ability.
Another route for the commercialization of autonomous driving, RoboTaxi, which is also regarded as the final application scenario of autonomous driving, has become a hot spot in the industry. In this regard, Xiao Jianxiong, founder and CEO of AutoX (Antu), shared his own views: only by achieving the same practicality of existing online ride-hailing vehicles, and completely removing the safety officer, unlimited purpose, and unlimited area of autonomous driving, is the real commercialization.
In addition, the "Chief Intellectual Officer Conference" also set up two roundtable forums: "Chip Challenges in the Era of Large Computing Power" and "How to Mature the Commercialization of Autonomous Driving".
Yang Yuxin, Chief Marketing Officer of Black Sesame Intelligence, Wang Ping, Executive President of Cambrian Xingge, Wang Kai, Director and CEO of Core Qing Technology, and Li Bo, Vice President of Lotus Technology and Head of Intelligent Driving Business Line, conducted a comprehensive discussion on chip issues.
2021 is called the first year of lidar on the car, accompanied by the automatic driving computing platform began to break through 1000 TPOS. This trend in Yang Yuxin's view, now the computing power has become an important indicator to judge the degree of intelligence of the car, the car companies hope to highlight the computing power value, so that the end user has more awareness of the automatic driving ability of the car company. The current computing power can theoretically meet the needs of L2+ and L3 autonomous driving systems, and the next focus is to make the scene and experience better.
He also added that "computing power stacking" is a necessary redundancy for subsequent technology upgrades, from the perspective of business logic and technology evolution, chip companies also need to help customers with smaller costs, higher system concentration, lower power consumption, to achieve better automatic driving functions, which is what chip companies have been working hard to promote, but also to promote the evolution of everyone's technology evolution and product routes.
Li Bo, who represents the demand side of the main engine factory, explains the meaning of hardware redundancy from another dimension. He proposed that software-defined cars, hardware-defined software ceilings, reserve enough computing power, reserve enough sensors, is to leave redundancy for the performance requirements of future intelligent driving systems. Otherwise, it's like the current application logic can run through the old phone, but it can't really run effectively.
Wang Ping also mentioned that at present, under the OTA trend, the car has shown a trend of gradual decoupling of software and hardware. Compared with hardware, software is easier to upgrade through OTA, prompting car companies to selectively pre-embed in computing power, even if this part is not used now.
In addition, Tesla, Xiaopeng and other car companies self-developed autonomous driving computing chips are also becoming a trend. Wang Kai, director and CEO of Core Qing Technology, said that this is because the shortage of chips has made the main engine factory pay more attention to the diversity of the supply chain and supply security, on the other hand, the high-computing chip has become the core competitiveness of the car company, and the supplier chip is becoming more and more difficult to meet the iteration speed, cost and performance requirements of the main engine factory.
But he also believes that this route faces many challenges: the threshold of autonomous driving chips is high, and once it takes a detour, it will face huge financial losses, which will also cause incongruity in planning. Vehicle-grade chips are different from consumer-grade chips, with higher requirements for performance, power consumption and reliability, and also to complete vehicle-level certification, longer cycles, greater investment, and need to recover the upfront costs through the application of a variety of vehicles, so it is necessary to launch a more inclusive and competitive product system to meet the needs of different car manufacturers.
In addition, the participants also answered the problem of chip shortage. The consensus is that the current cost of expansion is high, and chip manufacturers are blindly expanding production capacity without guaranteeing that there will be the same demand in the next few years. Although the current capacity has recovered from the epidemic, the demand that was suppressed last year has not yet been met, and the real solution may wait until next year.
In the second roundtable session on "Commercialization of Autonomous Driving", Zhou Xin, co-founder and chief product officer of Yishi Technology, Hao Jiannan, co-founder and chief architect of Tucson Future, Dong Jian, co-founder of Hongjing Intelligent Driving, VP of Kui Software Algorithm, and Dai Zhen, vice president of Heduo Technology, also had a heated discussion.
Zhou Xin, co-founder and chief product officer of Yishi Technology, and Hao Jianan, co-founder and chief architect of Tucson Future, believe that efficiency and cost are the premises for the commercialization of autonomous driving at the B-end: either to achieve higher efficiency than people, or to achieve fully unmanned autonomous driving. But to achieve the final business logic, not only a very high level of safety and reliability is required, but also the gradual improvement of regulations.
As a company that is facing both B-end and C-end users, Dong Jian, co-founder of Hongjing Intelligent Driving, said that the current landing speed is also faster than expected, and more mass production models will appear in one or two years. However, subject to legal and regulatory issues, most car companies will launch models with L3 level automatic driving experience, but developed according to the L2 + level regulatory system.
Dai Zhen, vice president of Heduo Technology, also gave a more specific time point for the C-end landing of automatic driving - it is expected that 2025 will be a key time node, when the mass production of autonomous driving technology, consumer acceptance, infrastructure and improvement of laws and regulations will gradually land.
AI x Science, the future has come
In the "AI x Science Forum" forum, Professor Xu Jinbo of the Toyota Institute of Computing Technology in Chicago, USA, made a report entitled "Protein Structure and Function Prediction". The report summarizes research advances in the field of protein mechanism prediction: artificial intelligence has upended the development of protein structure prediction and changed the research thinking of molecular biologists from sequence-based research to structure-based research. It also promotes structure-based drug discovery and design, improving the efficiency of protein de novo design.
On the other hand, Professor Xu also pointed out that there are still some problems that have not been fully solved in the current protein structure prediction, such as the interaction of proteins with other molecules, the impact of single-point mutations on protein structure and function, orphan protein structure prediction, and so on.
Subsequently, Dr. Guo Tiannan, Distinguished Researcher of Westlake University, Doctoral Supervisor, Director of Westlake Lab iMarker, and Founder of Westlake Omey, made a speech entitled "AI-Enabled Proteome Big Data Technology". Taking the team's recent research as an example, the speech demonstrated the value and application of proteomics in life sciences, applied artificial intelligence to proteomics, and combined with a large amount of clinical data to explore biomarkers, accelerate the latest progress of proteomics technology achievements in the field of oncology, and introduced cutting-edge information such as industrial transformation/landing in the field of AI-enabled proteomics big data technology.
At the "AI x Science Forum", Xia Ning, founder, chairman and CEO of Zhihua Technology, shared the relevant content with the theme of "AI-assisted chemical synthesis route design helps improve the efficiency of innovative drug research and development".
New drug research and development faces huge pain points of high cost, long time and low success rate. Zhihua Technology focuses on the design of chemical synthesis routes, and its independently developed algorithm is decomposed based on data learning and chemical knowledge, which solves the two major problems of interpretability and chemical reaction data. In addition to the reverse synthesis platform, research has been carried out in the fields of chemical process route design, chemical reaction conditions, by-product prediction analysis, and molecular pool generation. In the future, Zhihua Technology will continue to optimize to provide the diversity and feasibility of routes, use the failure response data from ELNs to avoid failures, and conduct multi-step strategy learning.
He Jingzhou, technical director of the natural language processing department of Baidu Shenzhen R&D Center and head of the propeller PaddleHelix biocomputing platform, delivered a keynote report entitled "Paddle Helix Empowers Biomedicine: Exploration and Application of AI Technology in the Field of Drug Research and Development". He shared the challenges and reflections of AI in the biomedical industry, as well as helping the biomedical industry, systematically presented the progress made based on pre-training technology and propeller Paddle Helix: compound characterization model GEM and protein PPI characterization model S2F.
He Jingzhou pointed out that AI has great potential in drug research and development, and in the future, pre-training will use massive unlabeled data for self-supervised learning, multi-task learning to enhance model generalization capabilities, and molecular spatial structure characteristics for model characterization, which can greatly improve AI production efficiency and reduce the threshold of drug development and production.
Wang Xuanze, founder and CEO of Creative Materials, shared the relevant content of "AI + Metal Materials: More Suitable for the Direction of Industrial Landing" in the "AI x Science Forum", and discussed the problems that may be encountered in the process of LANDING in the AI industry, including precision traps; industry barriers are not technology, traditional large factories transform into self-research; some problems in some toB fields; the nature of black boxes is serious, customers do not recognize, and they cannot replace key positions in the short term The algorithm effect is amazing but the landing is difficult and other wonderful views.
Wang Xuanze said that the industrialization of AI-enabled metal materials can effectively circumvent or solve the difficult problems in the above AI landing. On the other hand, high-end metal materials are an often overlooked market, and with industrial upgrading and strategic transformation, the demand for localization substitution is rapidly amplified. The main difficulty in the high-end metal field lies in the long R&D cycle and excessive R&D investment, so the research and development of new materials using AI to empower has become the optimal solution for overtaking in curves.
Ms. Wang Xiao'an, Founder and CEO of Brainland Technology, gave a speech entitled "AI-based Brain-Computer Technology Helps Broader Social Value and Mechanism Discovery".
Brain science aims to elucidate how and how the brain and nervous system work. The speech pointed out that with the breakthrough of artificial intelligence technology, the huge potential of brain science has once again been highly valued by the scientific community. Since 2019, the combination with AI has promoted the rapid development of brain magnetic imaging, brain-computer interface and other technologies at the application level, providing new solutions for the diagnosis and treatment of brain diseases, mental and sleep health management, entertainment interaction, safety production and other industries. In the future, the industry-university-research session will jointly discover more and more brain mechanisms and serve a wider range of people.
Dr. Wenbing Huang, Assistant Professor of the Intelligent Industry Research Institute (AIR) of Tsinghua University, delivered a speech entitled "GNN for Science: Graph Mechanics Networks" for the "AI×Science Forum". He introduced the application of artificial intelligence combined with the many-body problem in physics, and interpreted a new graph neural network - graph mechanics network GMN, which integrates the laws of physics into the construction of graph neural networks, and initially explores the advantages of data-driven and knowledge-driven combination. He also explains the physical and biomedical applications of GMN.
Dr. Huang pointed out that there are already more and more AI methods that shine in solving problems such as traditional natural sciences, and in the future, more attention can be paid to how to combine existing data-driven machine learning models with knowledge in basic science fields. Of course, there is still a relatively initial exploration.
The heart of the machine will upload a playback video on the B station in the future, and will also organize the content of the guest speech into text for release, welcome everyone to pay attention.
Playback viewing address: https://space.bilibili.com/73414544