"Across the side of the mountain, it is a peak, and the distance is different."

Using this poem to describe the current understanding of artificial intelligence is most appropriate.

After the rising tide of artificial intelligence, various “artificial intelligence” emerged in the medical field. In some medical conferences, wearable devices, automation systems, expert systems, intelligent equipment, the latest auxiliary diagnostic systems, and traditional CAD Technology is called artificial intelligence by companies or doctors.

What is artificial intelligence? Or what is the difference between artificial intelligence in the new era and artificial intelligence that emerged more than ten years ago? In terms of medical care, what kind of stage is the study? How long can this wave last? This is a problem that has been heard recently.

With these questions, the reporter reviewed some materials and interviewed Stanford University professor AI, Baidu Deep Learning Institute experts, medical AI company chief scientist, entrepreneur, Jilin University wisdom AI responsible professor and investment in medical AI The billion-dollar investors hope to get answers from them.

Parrot Intelligence and Crow Intelligence

UCLA, professor of statistics and computer science, director of the Center for Visual, Cognitive, Learning and Autonomous Robotics at the University of California, Los Angeles, has been in the visual search for an article "On Artificial Intelligence: Status Quo, Tasks, Architecture, and Unity | The Original Qingyuan" In the two metaphors, it is very interesting: "Parrot intelligence" and "Crow intelligence."

Parrots have a strong ability to imitate language. People often educate them several times. Parrots can repeat human phrases. This is similar to current data-driven chat bots.

Both can speak, but parrots and chatbots do not understand the context and semantics of speaking, nor can they relate the spoken words to objects, scenes, and characters in the physical world and society, and do not conform to cause and effect.

Crows are far more intelligent than parrots. They use tools to understand physical common sense and simple social common sense.

The above picture was taken by a Japanese researcher. The crow is wild. It must rely on its own observation, perception, knowledge, learning, reasoning, and execution to completely live independently.

In order to obtain the food in the nut, after various attempts, it found that putting the fruit on the road allowed the car to be rolled over to get the food (Figure b). However, the cars on the road are apt to lose their lives.

Then it was standing on the pole above the road. He hadn't studied big data before. It found that cars and people sometimes stopped near the intersection of red and green street lights.

At this time, it must further comprehend the complicated causal chain between traffic lights, zebra crossings, pedestrian lights, car stops, and pedestrian stops (Figures c, d, e). Eventually it found the rules and eventually got food.

In this process, crows are completely autonomous intelligence, fully autonomously aware, cognitive, reasoning, learning, and executing. There are also no millions of training data for him to learn. In this process, he also consumes very little energy (the human brain consumes about 10-25 watts, and the crow's brain is about 1% of humans. It has only 0.1-0.2 watts).

Parrot intelligence belongs to the data-driven intelligence that is now popular in applications. Although it is different from crow intelligence, it is promising in the vertical field of medicine. Crow intelligence belongs to artificial intelligence that people want.

Specifically, what research and application belongs to the category of artificial intelligence, Zhu Songchun gave a systematic induction, which covers six areas:

(1) Computer vision (for the time being, pattern recognition, image processing, etc. are included).

(2) Natural language understanding and communication (for the time being, speech recognition and synthesis are included, including dialogue).

(3) Cognition and reasoning (contains various physical and social common sense).

(4) Robotics (machinery, control, design, motion planning, mission planning, etc.).

(5) Game and ethics (multi-agent agents interaction, confrontation and cooperation, robots and social integration issues).

(6) Machine learning (various statistical modeling, analysis tools, and calculation methods).

Due to the decentralization of disciplines, most of the doctors, professors, and other professionals involved in related research often involve only one or more of these disciplines, or even focus on specific issues in a particular discipline for a long period of time.

For example, face recognition is a very small issue in the discipline of computer vision; deep learning is a popular genre in the discipline of machine learning.

In the field of medical applications, according to the arterial network, it is known that medical artificial intelligence products such as auxiliary diagnosis systems and image-assisted diagnosis systems currently seen in the market are cross-cooperation in various fields and combined with the results obtained through clinical practice, only in one area. Research is hard to achieve.

Three questions teach you to identify artificial intelligence companies

Wu Ren, a world-renowned computer game expert, believes there are three main reasons why this round of artificial intelligence can achieve unprecedented progress:

First, the accumulation of big data. In medicine, with the accumulation of electronic medical records and digital films, a large number of structured cases that can be used for research are preserved. This big data starts with at least 100,000 copies and can be used directly for model training data.

Second, the increase in computing power. In recent years, the improvement of computing power such as cloud computing and GPU has provided the basis for processing big data, shortening the time spent on training models, and shortening the training period to several days.

Third, the combination of big data and computing power enables researchers to quickly obtain and train algorithms that can be applied in time. Modern products that can be called artificial intelligence must have the participation of deep learning technologies. The previous artificial intelligence technology did not solve the problem well, and the emergence of deep learning technology only promoted this wave.

Wu Ren said that precisely because of the rise of this wave of artificial intelligence, it was determined that this wave would be continuous rather than intermittent. Judging from the current achievements, this claim is also being verified.

Based on these three aspects of common sense, we can use the following questions to identify the new era of artificial intelligence companies.

1. Where does the company's data come from, is it downloaded online or obtained from hospitals with high-quality labeling data?

2. Where does the company's algorithmic model come from, is it downloaded from the Internet and trained by others, or is it self-training model? If it is a self-training model, which is the training software and how strong is the computing power?

3. What is the difference between the self-training model and the open source model of others? There are several GPUs in the trained computer?

Easy-to-understand 3 criteria

In addition to the technical identification, the founder and CEO of Tuoma Shenwei also provided an easy-to-understand method for determining medical AI companies.

1. Has a core technical team that has long been engaged in the research of medical background artificial intelligence technology.

For example, Dr. Chen Yunqiang, chief scientist of Tuma Shenwei, is a biomedical engineering major at Tsinghua University. After graduating, he entered the Institute of Automation of the Chinese Academy of Sciences, the highest academic institution of Chinese natural sciences, specializing in artificial intelligence.

In 1998, he went to the University of Illinois at Urbana-Champaign with a Ph.D. degree in Electronics and Computer Engineering, under the tutelage of “Father of Computer Vision” Professor Huang Xiaotao.

Long-term development of medical image computer vision and artificial intelligence research and development in Siemens R & D global R & D center, this type of medical and artificial intelligence are complex talents with a profound background is very scarce. Can attract such top experts to join the company and represent one of the core competencies of Tuma Shenzhen.

2. Participate in academic conferences and exhibitions related to medical artificial intelligence at home and abroad and exchange achievements.

In the process of participating in the meeting, companies will participate in academic and scientific discussions, and they will present their own company and products in the exchange.

For example, at the North American Radiological Annual Conference, Tuma Shenwei made a debut of six major product systems and conducted in-depth exchanges with industry professionals at home and abroad.

At this exhibition, there are many friends and businessmen also participated in the exhibition and academic exchanges. Maybe we have a competitive relationship now and in the future, but we very much respect the research results of our friends. If a company is out of the artificial intelligence circle for a long time and just does some packaging publicity, it is doubtful.

3, the company needs to have the products of the hospital floor, as well as the doctor's recognition of the product.

Tuma Dewei has analyzed more than 50,000 cases of chest CT scans that have been widely recognized by doctors.

If it is only on the company's website to write that it belongs to an artificial intelligence company and there is no actual product, it is very difficult to finally obtain the endorsement of the final customer. This makes it difficult for the company to obtain long-term survival opportunities.

Zhong Hao concluded that those companies that have mastered the core technology, do practical things, and turn the needs of doctors into products that have landed on the ground have very good prospects for development.

Artificial intelligence must understand the physical world and its causal chain

Some experts have expressed that artificial intelligence can find some unknown connections with humans, rather than simply copying the functions of experts. Under different environmental conditions, the form of intelligence will be different.

Any intelligent machine must understand the physical world and its causal chains to adapt to this world.

Hessian Heterogeneous Song Jie said that the development direction of medical AI must not rely solely on human understanding of the disease's relevance and characteristics to allow computers to make judgments on diseases. Medicine cannot be perfectly structured, and it is necessary to know human understanding of nature. Perhaps only a few percent of the nature of nature, if we use this percentage of 'experience' to 'standardize' AI's understanding of nature, then there will be no breakthrough in the future of AI, and the development bottleneck will not exceed that of human doctors. The cognition. We also hope that AI can help us find more connections and the best treatment for diseases that we have not yet realized.

Various modern technologies may have good application possibilities in the medical field.

However, the problem solved by AI is the efficiency of human understanding of diseases. AI can find out the correlations and characteristics of diseases that humans need to discover for several years and find them out in a short time. The process may be several days, and the future may be several. Hours, this is the real ability of AI!

Conversely, at this stage, AI relies on the support of big data that humans already have, and what is claimed to be a "good model" based on very limited data can be said to be none of the true AI.

Various technologies may have medical application prospects, but it is not necessarily that they have to ride AI.

Some people think that medical artificial intelligence can automatically and semi-automatically diagnose diseases.

The system reads a large number of electronic medical records or medical knowledge, and then forms a set of knowledge maps, and has the ability to logically reason, make inferences based on the characteristics of the input patient's disease, and give a diagnosis basis.

The core is to systematically find the characteristics and laws of the disease and create rules by itself, rather than the characteristics that the researchers tell the system diseases.

And those systems that "write dead" rules are not artificial intelligence. For example, if the system writes all 10 features of the disease, only by satisfying these 10 features can the disease be diagnosed. These are obviously not artificial intelligence systems.

Deep learning is a sign of new era artificial intelligence

Zhu Songchun said that under the framework of probability statistics, many current deep learning methods belong to what I call “big data for small tasks”.

Designing a simple cost function Loss for a specific task, such as face recognition and object recognition

This method is also effective on some issues. However, as a result, this model cannot be generalized and interpreted.

Ye Weigang, founder of Delta Capital, said that the artificial intelligence in the new era must have deep learning capabilities. Looking at the medical artificial intelligence startup companies in China and the United States, most of them are companies with a computer background. They usually use some basic data to make mathematical models directly. Then the combination of calculation models and medical procedures is very much needed for industry integration.

As a capital partner, Delta is focusing on companies with bottom-level R&D technology and application-based artificial intelligence companies.

Application-based artificial intelligence enterprises can well solve the pain points in the industry under the premise of localization.

For example, the medical doctor Hui Ying of Dita Investment, based on the common deep learning algorithm framework, has made targeted developments for image special issues such as three-dimensional, large-scale, high-grayscale, and small data volume. The underlying framework is optimized so that artificial intelligence can have a strong degree of recognition in medical images, and establish barriers to the application of medical imaging.

The result is more important

Finally, a doctor who has long been engaged in artificial intelligence research has also given different conclusions. Doctors believe that as long as they can improve efficiency, reduce repetitive tasks, and provide better diagnostic accuracy, whatever the underlying technology, it can be Called artificial intelligence system, there is no need to have to worry about the source and name of the technology.

Watch & Apple Airtag Screen Protector

TPU Watch Screen Protector, Transparent ​Watch Screen Protector, Watch Screen Protector Case, Apple Watch Case, Apple Watch Case Protector, Apple Watch Protective Case

Shenzhen Jianjiantong Technology Co., Ltd. , https://www.jjttpucuttingplotter.com