The attitude is gradually becoming "close to the people". The five major AI trends in 2021 are highly anticipated

  World wave of technological innovation

  ◎Our reporter Liu Xia

  Artificial intelligence (AI) has become a “sweet and delicious” in the field of science and technology development in many countries. Governments and many large companies are not to be outdone, and are vying to spend heavily to support the development of this field. As a result, various innovations have sprung up like bamboo shoots after a rain. .

  In addition, the raging new crown pneumonia epidemic has forced us to further increase our reliance on technology, online activities and artificial intelligence.

Among them, artificial intelligence is particularly important for enterprises. It can realize personalized services on a large scale while meeting the ever-increasing experience needs of customers.

  The US "Forbes" biweekly website reported on March 15th for us the five most anticipated trends in the field of artificial intelligence in 2021.

These include the emergence of low-code/no-code tools, and becoming more and more "people-friendly", so that children can easily create their own artificial intelligence.

  Low-code/no-code tools

  Automated machine learning (AutoML) is not new. In 2020, Huawei’s PhD in machine learning will be recruited with an annual salary of one million. One of the research directions is AutoML.

  Machine learning is to allow algorithms to automatically find a set of rules from the data, thereby extracting relevant features in the data. With the development of machine learning, more and more parts of manual intervention are required, while AutoML is a comparison of machine learning models. The whole process from construction to application is automated.

  Although AutoML can build high-quality artificial intelligence models without solid data science knowledge, the low-code/no-code platform takes it to the next level-it can build the entire production-level artificial intelligence without in-depth programming knowledge Driven application.

  Last year, low-code/no-code tools suddenly emerged and swept the world, and the application fields are also varied. From building applications to enterprise-oriented vertical artificial intelligence solutions, this new force is expected to continue to exert strength this year.

  Statistics show that low-code/no-code tools will become the next frontline of the technology giants' battle. This is a market valued at 13.2 billion U.S. dollars, and its total value is expected to further increase to 45.5 billion U.S. dollars by 2025.

  The Honeycode platform released by Amazon in June 2020 is the best proof. Honeycode is a non-code development environment similar to a spreadsheet interface and is known as the "gospel" for product managers.

  Advanced pre-trained language model

  "The Representation of Bidirectional Encoders from Transformers" (BERT) is a new language model developed and released by Google at the end of 2018.

As a rookie in the field of natural language processing (NLP), BERT has become a master of the major advances in NLP in the past few years. It shocked four competitors and broke the highest record of 11 NLP tests, even surpassing human performance. .

  In recent years, pre-trained language models similar to the BERT model (such as question answering, named entity recognition, natural language inference, text classification, etc.) have played an important role in many natural language processing tasks.

  These pre-trained language models are very powerful and have completely changed the language translation, understanding, and summary, etc., but these models are very expensive and very time-consuming to train.

  The good news is that advanced pre-training models can give birth to a new generation of efficient and extremely easy-to-build artificial intelligence services.

  GPT-3 is the leader among them!

It is a natural language processing model built by OpenAI with a huge amount of money. It has a large parameter volume of 175 billion and is the strongest AI model in the NLP field.

Since it was first launched in May last year, with its amazing text generation capabilities, GPT-3 has remained popular on major media platforms.

It can not only answer questions, write articles, write poems, translate articles, but also generate codes, do mathematical reasoning, data analysis, draw charts and make resumes, and even play games, and the effect is surprisingly good.

  Synthetic content generation

  Algorithm innovation in the field of artificial intelligence does not just appear in NLP.

Generative Adversarial Networks (GANs) have also seen a lot of innovations, demonstrating the extraordinary achievements of scientists in creating art and fake images.

  GANs were first proposed by Ian Goodfellow, an AI scholar at the University of Montreal, Canada, and their training and adjustments are also complicated because they require a large number of data sets for training.

  But the innovation of scientists has greatly reduced the amount of data required to create GANs.

For example, NVIDIA Corporation of the United States demonstrated a new method to enhance the efficiency of training GANs, which requires less data than previous methods.

This allows GANs to be widely used in many fields, from medical applications (such as synthesizing cancer histological images) to deeper "Deep Fake".

  "Deep fake" is a high-energy black technology that uses the latest artificial intelligence technology to allow ordinary people to edit some videos through the computer, and the human face in the video can become anyone's face.

"The so-called success is also Xiaohe, and the failure is also Xiaohe", while the video "changing face" has aroused huge attention, it also caused huge controversy.

Just five days after going online, this black technology was spurned by the entire network, and then blocked globally.

  Artificial intelligence for children

  With the popularity of low-code tools, AI creators also show the characteristics of a younger age.

Now, an elementary and middle school student can create artificial intelligence for his own use-from categorizing text to drawing images.

American high schools have already opened artificial intelligence courses, and junior high schools are not far behind.

  For example, at the 2020 Synopsys Science Fair in Silicon Valley, 31% of award-winning software projects used artificial intelligence in their innovation.

Even more impressive is that 27% of these artificial intelligences were created by students in grades 6-8.

One of the winners is an eighth-grade student who created a convolutional neural network that can detect diabetic retinopathy through eye scans.

  Machine learning operations

  Machine learning operations (MLOps) is a relatively new concept in the field of artificial intelligence, involving the best management data scientists and operators in order to effectively develop, deploy, and monitor models.

  In 2020, due to the raging new crown epidemic, huge changes in operational workflows, inventory management, and traffic patterns have caused many artificial intelligence to exhibit unexpected behaviors. This is called drift-the input data does not match the expectations of artificial intelligence training.

  Although companies that deploy machine learning in production have faced many challenges such as drift before, the new crown epidemic has increased the demand for MLOps.

Coincidentally, with the implementation of privacy regulations such as the California Consumer Privacy Act of 2018, companies that operate on customer data increasingly need governance and risk management.

According to data, the market size of MLOps is expected to reach US$4 billion by 2025.

  These are not all new trends in the field of artificial intelligence, but they deserve our attention because they highlight three important aspects.

First of all, there are more and more applications of artificial intelligence in the real world, as evidenced by the problems caused by the new crown epidemic and the growth of MLOps.

Secondly, relevant people continue to innovate in this field, just as BERT and GANs followed one after another.

Finally, the threshold for creating artificial intelligence is getting lower and lower, laying a solid foundation for it to "fly into the homes of ordinary people."

  The ideals and future of artificial intelligence are always beautiful, but despite the many innovations mentioned above, we still need to promote and guide its development in a down-to-earth manner so that it can better benefit mankind.