In the last year, I’ve led over 20 AI workshops teaching engineers the basics in AI systems and getting them to the point where they can implement the start of the art systems available. It is a fantastic experience to teach deep learning in China and help build the skills of 1000’s of engineers. While I might have been the teacher, I ended up learning a lot myself about the state of china’s AI field.
Here are the top five things I picked up while teaching in China.
Chinese engineers have serious talent and know-how
I feel like there is this idea that China the land of copycats and that many Chinese companies and engineers have no skill outside of copying others. This couldn’t be further from the truth. I’ve spent the last five years in China, and after all these years I’m still meeting new amazing and talented engineers every day.
While in the past many left China to live in Europe and the USA after studying in college, nowadays they are choosing to come back to China or stay to study at one of China’s amazing technical universities like Tsing Hua. When I first started teaching classes, I made my lesson plan such that it would start slow. After my first class, I had to redo everything about it. The Chinese engineers learned fast and demanded more examples than I had available! Eventually, I was teaching state-of-the-art papers that just came out!
Receive weekly AI research papers, resources & trends. Subscribe to AI Scholar for FREE
If you are interested in my classes you can get a super distilled high-level overview from my article series Becoming a Machine Learning Engineer
The amount of data available for training is enormous
If AI is the new electricity, then data is the coal that fuels the power plants. Data is so essential I see it as one of the easiest ways to boost performance on deep learning algorithms and even write about it in my Deep Learning Cheat Sheet.
AI’s enormous importance hasn’t convinced everyone. Europe recently passed the General Data Protection Regulation (GDPR), and the US has aspirations to join them with similar data regulations. These regulations while written to help the consumer have the effect of significantly reducing the capacity of companies and individuals to build AI systems that rely on free access to data.
In China, there are very few restrictions on data, and that has led to a data explosion. This is a problem that is only going to get worse as more and more Chinese companies start to leverage the data that is generated by 1.4 billion people.
Research takes a back seat to Implementation
The Chinese engineers I taught were equal in two things. Skill and the desire to get straight to implementation. It seemed everyone had an idea of a product they wanted to build. A few of those eager students have gone on to start companies with what they learned in my classes using some bare-bones algorithms and datasets, but their drive to implement meant they were ahead of the curve and have early successes.
While my students seem to have boundless energy for implementing AI systems into their products, very few students seemed excited about the idea of actually working on novel deep learning systems. It’s likely because of the vast amount of money flowing into the AI companies that many are not as interested in hard research.
The Chinese government takes AI very seriously
Months into my training sessions I started to notice a significant spike in the number of students in my classes that were coming from companies. It turns out that the government was giving grants to help pay for training in AI implementation. That made me look into other ways the Chinese government is investing in AI.
Turns out they are spending billions of dollars to boost the nation’s AI capability and have made it a central part of their made in China 2025 plan. The AI industry under this massive investment has grown 67 percent over the past year and produced more patents and research papers than in the US. All of this with only about a fifth of America’s talent pool. I wouldn’t be surprised if they do end up being the world leaders in AI by 2030.
There is still a long way to go for information to disseminate
Lots of info on deep learning and AI is still only accessible in English, while that was not a problem for the students that I had in my classes, they did spend a lot of time recounting stories of friends whose lack of English skills made it difficult for them to get into the field and start implementing.
As more and more information crosses the language barrier, we can expect more interesting AI applications coming from those left out at the beginning.