BMW headquarters, Munich, Germany.
Andriy Burkov
These are the best posts from Andriy Burkov.
48 viral posts with 126,968 likes, 4,110 comments, and 2,679 shares.
29 image posts, 0 carousel posts, 3 video posts, 16 text posts.
👉 Go deeper on Andriy Burkov's LinkedIn with the ContentIn Chrome extension 👈
Best Posts by Andriy Burkov on LinkedIn
The PULSE algorithm takes pixelated faces and turns them into high-resolution images. When it takes a pixelized Obama face it outputs a white face. This is an example of bias in AI: the situation when a machine learning model was trained on the data that has a high imbalance of some class (in this case, white people).
People, including many scientists, tend to think that AI is right because its impartial and based on data. However, the AI is only right when trained on balanced data. What is balanced and how to ensure the right balance, it's a huge unanswered question.
Image: Twitter / @Chicken3gg
PULSE on GitHub: https://lnkd.in/efde43e
People, including many scientists, tend to think that AI is right because its impartial and based on data. However, the AI is only right when trained on balanced data. What is balanced and how to ensure the right balance, it's a huge unanswered question.
Image: Twitter / @Chicken3gg
PULSE on GitHub: https://lnkd.in/efde43e
PhD has lost its cool. It's no longer needed to have a PhD to be considered a highly professional data scientist or a machine learning engineer. Don't do a PhD unless what you want is to teach or do scientific research.
Your employer has no legal right to ask you to do that, but I'm not your employer so I can tell you that: if you want to progress in your career as a data scientist, you should constantly learn new things. In the transport, in the evening, on the weekends. Don't wait for your employer to allocate a paid time slot for your learning. If you want to grow in what you do, *you* have to drive your growth. Don't fool yourself: no paid training once a year or a conference in San Francisco or Seattle will help you to become great in this craft.
Your employer said nothing.
Your employer said nothing.
In a typical machine learning project:
60% is data discovery/transformation/creation;
30% is coding not related to machine learning;
10% is actual machine learning.
60% is data discovery/transformation/creation;
30% is coding not related to machine learning;
10% is actual machine learning.
Treat your time after work as your second job where you are the project. Invest your time and money to learn a new thing, try a new activity, become more healthy, do something that gives you a sense of accomplishment about yourself.
You work hard for someone else's business. Work as hard for the business of your own well-being.
You work hard for someone else's business. Work as hard for the business of your own well-being.
Did you ever wonder why companies pay Google for the keyword search of their name if they are already on the top of the search results?
If you did, then here's why. They bid the highest price for the click so that their competitor (say, Nike) cannot place their ad higher than the organic link.
This is how Google made everyone pay.
If you did, then here's why. They bid the highest price for the click so that their competitor (say, Nike) cannot place their ad higher than the organic link.
This is how Google made everyone pay.
Finland offers free online Artificial Intelligence course to anyone, anywhere:
“Helsinki University and tech strategy firm Reaktor say they want to make Finland the world's most educated country in the field of artificial intelligence.“
“The academic and business partners say they want Finland to become forerunners in AI, and have developed an online course covering the quickly-growing technology open to anyone, free of charge.“
Building AI systems together as a people is definitely more fun than drinking vodka!
Source: https://lnkd.in/gj_mMfK
“Helsinki University and tech strategy firm Reaktor say they want to make Finland the world's most educated country in the field of artificial intelligence.“
“The academic and business partners say they want Finland to become forerunners in AI, and have developed an online course covering the quickly-growing technology open to anyone, free of charge.“
Building AI systems together as a people is definitely more fun than drinking vodka!
Source: https://lnkd.in/gj_mMfK
Each time I see that a job description requires 9 years of experience with something, I'm curious how they came to this number? How did the discussion happen? “Do you think that 8 years of experience is enough?“ “No, we need at least 9 years!“ or “Let's ask for 10 years of experience?“ “No, I think that 9 years is fine!“
Something like that?
Something like that?
Ok, that happened. My book is THE best-selling book on AI. Not the best-selling technical book, not the book on ML, the best selling book period. Can't remember when last time I was so excited!
I address this message to all data science, engineering and database folks. I think that in 2018 it's time to start not being afraid to say “data“. There's no big data and small data anymore. There's data, and you, as a professional, have to be able to store it, transform it and analyze it, no matter its size.
If you only comfortable to work with a small amount of data and you are in 2018, then you have to admit it: you need to improve your skills.
Let 2018 be the first year when we again can hire data engineers and not big data engineers, data analysts and not big data analysts, data architects and not big data architects, database developers and not big data developers.
I also wish that in 2018, scientists will not be afraid to call themselves research scientists or applied scientists and not data scientists.
Data is everywhere, it can be big, it can be small, it can be slow, it can be fast, it can be simple, it can be complex, it can be complete or incomplete, it can be clean, it can be noisy. In 2018, you are supposed to be able to deal with it.
If you agree with me, like or share this message. Let's get our self-respect and professionalism back!
If you only comfortable to work with a small amount of data and you are in 2018, then you have to admit it: you need to improve your skills.
Let 2018 be the first year when we again can hire data engineers and not big data engineers, data analysts and not big data analysts, data architects and not big data architects, database developers and not big data developers.
I also wish that in 2018, scientists will not be afraid to call themselves research scientists or applied scientists and not data scientists.
Data is everywhere, it can be big, it can be small, it can be slow, it can be fast, it can be simple, it can be complex, it can be complete or incomplete, it can be clean, it can be noisy. In 2018, you are supposed to be able to deal with it.
If you agree with me, like or share this message. Let's get our self-respect and professionalism back!
I see many people put their Coursera and other MOOC courses to the education section of their LinkedIn profiles like this:
—Stanford University
Deep Learning Specialization (on Coursera)
2017-Present
—Massachusetts Institute of Technology
Deep Learning with Python (on DataCamp)
2017
While it looks cool at first sight, it will most probably not be appreciated as a technique to make the education look better than it is by the recruiters and hiring managers.
Coursera courses are a very important asset in your profile, but they have to be called what they are: online virtual introductory courses. They don't give the same depth of knowledge as a semester spent in Stanford/MIT.
—Stanford University
Deep Learning Specialization (on Coursera)
2017-Present
—Massachusetts Institute of Technology
Deep Learning with Python (on DataCamp)
2017
While it looks cool at first sight, it will most probably not be appreciated as a technique to make the education look better than it is by the recruiters and hiring managers.
Coursera courses are a very important asset in your profile, but they have to be called what they are: online virtual introductory courses. They don't give the same depth of knowledge as a semester spent in Stanford/MIT.
The difference between machine learning research and machine learning practice is that in research, we spend months trying hard to design an algorithm complex enough to gain 0.3% improvement over the previous best algorithm.
In practice, we use a simple algorithm and spend one week annotating more data to gain a 3% improvement over the previous week's result.
In practice, we use a simple algorithm and spend one week annotating more data to gain a 3% improvement over the previous week's result.
One nuclear physicist told a story. He changed the domain from nuclear physics to software development. He says: “Working in IT is quite cool: you just download the needed library, then write code and google solutions in case of problems. In nuclear physics, the needed thing will be built in 20 years and the only document you can google on the topic was written by you.“
— Forget everything you learned in college. You won't need it working here.
— But I never went to college.
— Well then, I'm sorry. You are underqualified to work here.
— But I never went to college.
— Well then, I'm sorry. You are underqualified to work here.
A physicist, engineer and a statistician are out hunting. Suddenly, a deer appears 50 yards away.
The physicist does some basic ballistic calculations, assuming a vacuum, lifts his rifle to a specific angle and shoots. The bullet lands 5 yards short.
The engineer adds a fudge factor for air resistance, lifts his rifle slightly higher, and shoots. The bullet lands 5 yards long.
The statistician yells: “We got him!“
The physicist does some basic ballistic calculations, assuming a vacuum, lifts his rifle to a specific angle and shoots. The bullet lands 5 yards short.
The engineer adds a fudge factor for air resistance, lifts his rifle slightly higher, and shoots. The bullet lands 5 yards long.
The statistician yells: “We got him!“
The most popular languages on the Internet.
1 English 54%
2 Russian 6.0%
3 German 5.9%
4 Spanish 4.9%
5 French 4.0%
6 Japanese 3.4%
7 Portuguese 2.9%
8 Italian 2.3%
9 Persian 2.0%
10 Polish 1.8%
11 Chinese 1.7%
Source: W3Techs
I always say to my kids that you must know English just because it's almost guaranteed that you will *only* find in English the information you need.
1 English 54%
2 Russian 6.0%
3 German 5.9%
4 Spanish 4.9%
5 French 4.0%
6 Japanese 3.4%
7 Portuguese 2.9%
8 Italian 2.3%
9 Persian 2.0%
10 Polish 1.8%
11 Chinese 1.7%
Source: W3Techs
I always say to my kids that you must know English just because it's almost guaranteed that you will *only* find in English the information you need.
There's no AI bubble. There's an awesome technology and a bubble of people not having a clue about what to do with this technology.
Learn math and the machine learning basics and you will escape the bubble.
Learn math and the machine learning basics and you will escape the bubble.
People don't like Mark Zuckerberg, and probably there are reasons for that. But that witch hunt that is happening right now is beyond any reason.
They say that he has to stop political advertising. Did they hear about the freedom of speech? Who will decide what's political and what's not?
They say that he has to check ads for truth. Did they hear about the presumption of innocence and that only the court has the right to decide what is the truth?
America, what's happening with you? There are only two pillars of democracy: freedom of speech and presumption of innocence. Without these two, anything else doesn't matter.
They say that he has to stop political advertising. Did they hear about the freedom of speech? Who will decide what's political and what's not?
They say that he has to check ads for truth. Did they hear about the presumption of innocence and that only the court has the right to decide what is the truth?
America, what's happening with you? There are only two pillars of democracy: freedom of speech and presumption of innocence. Without these two, anything else doesn't matter.
I'm thrilled to announce that my new book, Machine Learning Engineering, was just released and is now available on Amazon and Leanpub, as both a paperback edition and an e-book!
You can find more information on the release in the below article. I'll be in the comments to answer all the questions.
#machinelearning #artificialintelligence #engineering
You can find more information on the release in the below article. I'll be in the comments to answer all the questions.
#machinelearning #artificialintelligence #engineering
Just compiled the first draft of the Machine Learning Engineering book. It will be 300 pages. One can now roughly estimate the timeshare of various activities in a machine learning projects: ~100/400 pages for the actual machine learning and ~300/400 pages for everything else.
If the stars align as expected, the book will be released on September 1st🤞.
If the stars align as expected, the book will be released on September 1st🤞.
