Reporting by XinZhiyuan
Editor: Good sleepy Yuan Xie
Wu Enda, one of the most authoritative scholars in the field of contemporary artificial intelligence, announced on his Twitter on the morning of February 8, 2022 that the new crown test result was positive, but the symptoms were mild.
Beijing time, at 6:00 a.m. on February 8, 2022, Ng Enda tested positive for the new crown virus.
Ng said that because he had been vaccinated and boosters, his current symptoms were similar to those of a mild flu. Thank you to the people working on vaccines around the world.
On the first day of the Chinese New Year, he also tweeted to wish everyone a happy Year of the Tiger.
There are still less than 3 months to celebrate the 46th birthday, I hope the big guy has a good rest and a speedy recovery.
Achievements at a glance
Ng is undoubtedly one of the most authoritative scholars in the field of contemporary artificial intelligence and machine learning, and he is also quite successful in business.
Ng is a visiting professor in Stanford's Department of Computer Science and Electrical Engineering, and was former director of the Stanford Artificial Intelligence Laboratory.
Ng's ideal is to make a high-quality, free education accessible to everyone in the world. Together with Daphne Kohler, a sister and big cow in the world of machine learning and author of Probabilistic Graphical Models: Principles and Techniques, he founded Coursera, an online education platform.
Ng was born in London, England in 1976. His parents are immigrants from Hong Kong. Growing up, he spent time in Hong Kong and Singapore before graduating from Raffles College Singapore in 1992.
In 1997, he received a triple-major in computer science, statistics, and economics from Carniki Mellon University in Pittsburgh, Pennsylvania. Between 1996 and 1998, he conducted reinforcement learning, model selection, and feature selection at AT&T Bell Labs.
In 1998, Ng received her master's degree from the Massachusetts Institute of Technology in Cambridge, Massachusetts. At MIT, he built the first publicly available, self-indexed web search engine for research papers on the web (it was the predecessor of CitesSeer/ResearchIndex, but focused on machine learning).
Fun Picture: "When you see the following title tag, you know that the film and television products are great: 20th Century Fox, Paramount, Warner Bros., Ng Enda Smile"
In 2011, Ng created the Google Brain project at Google to develop ultra-large-scale artificial neural networks through distributed cluster calculators.
On May 16, 2014, Ng joined Baidu to take charge of the "Baidu Brain" program and served as the chief scientist of Baidu. On March 20, 2017, Ng announced his resignation from Baidu.
In December 2017, Ng announced the establishment of an artificial intelligence company Landing.ai as the company's CHIEF EXECUTIVE OFFICER.
Fun Map: "The literary image of AI is the terminator, and the real image is the Ng Enda Open Class"
As a teacher, he holds a record: In the "Machine Learning" course at Stanford University in the fall semester of 2013-1014, more than 800 students took this course taught by Ng. This was the most popular course in Stanford history.
There are no classrooms to accommodate, so many people are watching class footage at home. But this graduate program in computer science is much more difficult than the open course of the same name on Coursera, which in his own words is "this (compared to Coursera) can be said to be two courses."
Professor Wu's open class golden sentence: "Don't be afraid if you don't understand it"
He lectures on machine learning in Stanford Open Classes and Coursera, which is extremely effective and very popular in the industry and the general public.
Funny: "Girlfriend: You don't cry when you look at the Titanic!" Unbelievable! Do you have feelings or not! Have you ever cried! AI student: Yes, when Ng Enda wrote a thank you subtitle at the end of the open class."
Ng's machine learning course at Coursera averaged 4.9 points. Coursera's courses are rated out of 5, most of the open courses are between 4-4.5, and few courses that can achieve 4.9 points, and this course has nearly 50,000 people rated. According to Freecodecamp, this is one of the most popular online courses in machine learning.
There are relatively few high-numbered content in Ng Enda's open courses, and it is more friendly to the public in similar open courses. He explains why: "The reason why this course doesn't use too much math is because it has a wide audience, so intuitive explanations give everyone the confidence to keep learning."
Fun map: "Ng Enda's open class silently blocked the high number firepower of calculus, line generation, statistics, and probability theory for AI Xinding, so that students can sleep peacefully."
80% of the data + 20% of the model = better machine learning
Whether advances in machine learning are brought about by models or data may be a century of debate.
Ng's idea is that 80% of a machine learning team's work should be focused on data preparation, and ensuring data quality is the most important job.
「AI = Data + Code」
When something goes wrong, most teams instinctively try to improve the code. But for many practical applications, it's more effective to focus on improving the data.
Ng believes that if there is more emphasis on data-centric rather than model-centric, then machine learning will evolve rapidly.
We all know Google's BERT, OpenAI's GPT-3. However, these amazing models only solve 20% of business problems. The remaining 80% is the quality of the data.
What is MLOps?
MLOps, a combination of Machine Learning and Operations, is a subset of ModelOps.
It is a practice of collaboration and communication between data scientists and operations professionals to help manage the lifecycle of machine learning tasks.
Similar to DevOps or DataOps approaches, MLOps wanted to increase automation and improve the quality of the production ML, while also focusing on business and regulatory requirements.
For example, when deploying AI in a data-starved application scenario, such as an agricultural scenario, you can't expect to have a million tractors collecting data for yourself.
Based on MLOps, Ng also made several suggestions:
The most important task of MLOps is to provide high-quality data.
Label consistency is also important. Checking whether labels have clear boundaries that govern themselves, even if the definition of labels is good, lack of consistency can lead to poor model performance.
It is better to systematically improve the data quality on the baseline model than to pursue the latest model with low-quality data.
If an error occurs during training, a data-centric approach should be taken.
If data-centric, for smaller data sets (
When working with smaller data sets, tools and services that improve data quality are critical.
Ten years ago, when communities began embracing deep learning, we didn't know how many tens of thousands of novel inventions and how many research papers would be needed to get to where we are today.
But the incomprehension of the past has long since vanished.
"People started questioning TensorFlow and other frameworks that laid the groundwork. And right now, I think there are tens of thousands of ideas to be invented at any time when it comes to thinking about MLOps and data-centric AI."
At the end of the day, any machine learning business has to care about the needs of its customers for the product, and everything is business-related.
When building mlOps teams, Ng recommends a solid principle:
Requires teams to take a long-term, hard look at it to ensure that high-quality data is consistently produced throughout the product lifecycle.
"Even though the term MLOps doesn't appear in job descriptions, I think MLOps is still an important skill that people need to learn right now."
Resources:
https://twitter.com/AndrewYNg/status/1490808144267673601