Top 10 Quora Machine Learning Writers and Their Best Advice
Top Quora machine learning writers give their advice on pursuing a career in the field, academic research, and selecting and using appropriate technologies.
By Matthew Mayo.
This list is based on Quora ranking of Machine Learning writers.
1. Yoshua Bengio, Machine Learning Researcher, Professor @ U. Montreal. 68379 views
Selected answer to: Does Yoshua Bengio prefer to use Restricted Boltzmann Machines or (denoising) Autoencoders as building blocks for deep networks? And why?
Scientific research is not about "preferring" one algorithm over another. It's about *understanding*. I am interested in understanding Boltzmann machines and I am interested in understanding all kinds of auto-encoders. The world of unsupervised learning is wide open. No clear winner but many interesting questions. As far as unsupervised pre-training goes, denoising auto-encoders are somewhat easier to train and use than RBMs, but give about the same results, so I would go for the former if I had to make a choice.
2. Yisong Yue, Machine Learning Researcher. 61209 views
Selected answer to: Is academic CS research really valuable?
Perhaps you should review the academic papers on deep learning? For a long time, many people thought those papers weren't that valuable. And now, deep learning and large-scale machine learning in general is fast becoming a billion dollar industry. It is absolutely the case that academic research set the stage for deep learning to really flourish in industry.
3. Gary Simon, Professor of Statistics (retired), Stern School at NYU. 50098 views
Selected answer to: If an event has a 10% chance of happening, why is it not guaranteed to happen in 10 tries?
Let's assume that the tries are independent and that the event probability does not change.
A parallel question is this: If a coin has a 50% change of coming up heads, why is it not guaranteed to happen in two tries?
It's easy to see through this!
4. Xavier Amatriain, VP of Engineering at Quora. 39980 views
Selected answer to: How do I learn machine learning?
My main point is that machine learning is both about breadth as depth. You are expected to know the basics of the most important algorithms (see my answer to What are the top 10 data mining or machine learning algorithms?). On the other hand, you are also expected to understand low-level complicated details of algorithms and their implementation details. I think the approach I am describing addresses both these dimensions and I have seen it work.
5. William Chen, Data Scientist at Quora. 30659 views
Selected answer to: What are some good resources to practice SQL for a data science interview?
I would do a review of SQLZoo by going through all of their exercises. They have an extremely convenient interface where you can write your SQL queries directly on their website, letting you run arbitrary SQL queries on the provided tables. They also will provide answers for you (Where can I find answers to SQLZoo exercises?).
If there's one particular concept you should be extremely familiar with, review your JOINS - as this is commonly asked and is easy to mess up.
6. Jeff Hammerbacher, Curious. 29996 views
Selected answer to: What are the top 10 data mining or machine learning algorithms?
One potential answer to this question comes from the Analytics 1305 [2] documentation:
7. Boxun Zhang, Data Scientist at Spotify; PhD in Computer Science. 26049 views
Selected answer to: To be a data scientist in a tech company (Google, Microsoft, Facebook, etc.), how well do I need to know machine learning algorithms?
As John L. Miller and John Eysman pointed out in their answers, it is rare that data scientists need to implement machine learning algorithms from scratch.
However, I would like to emphasize the importance of actually knowing how machine learning algorithms work, particularly the limitations of those algorithms. For example:
Selected answer to: Is a Ph.D. necessary for a job in Machine Learning?
Probably not. i think 10 years ago, this was true. Machine Learning was very new and those of us who understood it were trying to pioneer it's use in companies. These days, I meet a lot of young people who have a solid understanding of the basics and can do quality work.
That being said, as with advanced computer science, if you want to do truly innovative algorithm work--like create the next PageRank algorithm--then a PhD will expose you to the entire field and you will have the time and freedom to invent something new that could be potentially revolutionary.
9. Matthew Lai, CS MSc student at Imperial College. Electrical Engineer. Pilot. 20505 views
Selected answer to: What's the "Hello, World!" program of machine learning?
Randomly generate a bunch of points in the square (-1, -1)-(-1,1)-(1,-1)-(1,1).
Classify whether they are in the unit circle centered at the origin.
It will catch very bad mistakes (opposite gradient direction, etc), and doesn't require accessing any dataset.
10. Sean Owen. 20318 views
Selected answer to: Is it better for a data scientist to work for a big company or startup?
I'm afraid that data quality is a problem everywhere, and data engineering is most of the work. I prefer doing a bit of everything myself, but it sounds like you want to focus more narrowly on the modeling. A big company might be a better fit. However I'd have this conversation with your current company first to see if you can shift your responsibilities to what you want.
Bio: Matthew Mayo is a computer science graduate student currently working on his thesis parallelizing machine learning algorithms. He is also a student of data mining, a data enthusiast, and an aspiring machine learning scientist.
Related:
This list is based on Quora ranking of Machine Learning writers.
1. Yoshua Bengio, Machine Learning Researcher, Professor @ U. Montreal. 68379 views
Selected answer to: Does Yoshua Bengio prefer to use Restricted Boltzmann Machines or (denoising) Autoencoders as building blocks for deep networks? And why?
Scientific research is not about "preferring" one algorithm over another. It's about *understanding*. I am interested in understanding Boltzmann machines and I am interested in understanding all kinds of auto-encoders. The world of unsupervised learning is wide open. No clear winner but many interesting questions. As far as unsupervised pre-training goes, denoising auto-encoders are somewhat easier to train and use than RBMs, but give about the same results, so I would go for the former if I had to make a choice.
2. Yisong Yue, Machine Learning Researcher. 61209 views
Selected answer to: Is academic CS research really valuable?
Perhaps you should review the academic papers on deep learning? For a long time, many people thought those papers weren't that valuable. And now, deep learning and large-scale machine learning in general is fast becoming a billion dollar industry. It is absolutely the case that academic research set the stage for deep learning to really flourish in industry.
3. Gary Simon, Professor of Statistics (retired), Stern School at NYU. 50098 views
Selected answer to: If an event has a 10% chance of happening, why is it not guaranteed to happen in 10 tries?
Let's assume that the tries are independent and that the event probability does not change.
A parallel question is this: If a coin has a 50% change of coming up heads, why is it not guaranteed to happen in two tries?
It's easy to see through this!
4. Xavier Amatriain, VP of Engineering at Quora. 39980 views
Selected answer to: How do I learn machine learning?
My main point is that machine learning is both about breadth as depth. You are expected to know the basics of the most important algorithms (see my answer to What are the top 10 data mining or machine learning algorithms?). On the other hand, you are also expected to understand low-level complicated details of algorithms and their implementation details. I think the approach I am describing addresses both these dimensions and I have seen it work.
5. William Chen, Data Scientist at Quora. 30659 views
Selected answer to: What are some good resources to practice SQL for a data science interview?
I would do a review of SQLZoo by going through all of their exercises. They have an extremely convenient interface where you can write your SQL queries directly on their website, letting you run arbitrary SQL queries on the provided tables. They also will provide answers for you (Where can I find answers to SQLZoo exercises?).
If there's one particular concept you should be extremely familiar with, review your JOINS - as this is commonly asked and is easy to mess up.
6. Jeff Hammerbacher, Curious. 29996 views
Selected answer to: What are the top 10 data mining or machine learning algorithms?
One potential answer to this question comes from the Analytics 1305 [2] documentation:
- Kernel Density Estimation and Non-parametric Bayes Classifier
- K-Means
- Kernel Principal Components Analysis
- Linear Regression
- Neighbors (Nearest, Farthest, Range, k, Classification)
- Non-Negative Matrix Factorization
- Support Vector Machines
- Dimensionality Reduction
- Fast Singular Value Decomposition
- Decision Tree
- Bootstrapped SVM
7. Boxun Zhang, Data Scientist at Spotify; PhD in Computer Science. 26049 views
Selected answer to: To be a data scientist in a tech company (Google, Microsoft, Facebook, etc.), how well do I need to know machine learning algorithms?
As John L. Miller and John Eysman pointed out in their answers, it is rare that data scientists need to implement machine learning algorithms from scratch.
However, I would like to emphasize the importance of actually knowing how machine learning algorithms work, particularly the limitations of those algorithms. For example:
- Is feature engineering relevant at all for Random Forests?
- What kind of tree algorithm is the best for implementing a random forest and why?
- What is an intuitive explanation of gradient descent?
Selected answer to: Is a Ph.D. necessary for a job in Machine Learning?
Probably not. i think 10 years ago, this was true. Machine Learning was very new and those of us who understood it were trying to pioneer it's use in companies. These days, I meet a lot of young people who have a solid understanding of the basics and can do quality work.
That being said, as with advanced computer science, if you want to do truly innovative algorithm work--like create the next PageRank algorithm--then a PhD will expose you to the entire field and you will have the time and freedom to invent something new that could be potentially revolutionary.
9. Matthew Lai, CS MSc student at Imperial College. Electrical Engineer. Pilot. 20505 views
Selected answer to: What's the "Hello, World!" program of machine learning?
Randomly generate a bunch of points in the square (-1, -1)-(-1,1)-(1,-1)-(1,1).
Classify whether they are in the unit circle centered at the origin.
It will catch very bad mistakes (opposite gradient direction, etc), and doesn't require accessing any dataset.
10. Sean Owen. 20318 views
Selected answer to: Is it better for a data scientist to work for a big company or startup?
I'm afraid that data quality is a problem everywhere, and data engineering is most of the work. I prefer doing a bit of everything myself, but it sounds like you want to focus more narrowly on the modeling. A big company might be a better fit. However I'd have this conversation with your current company first to see if you can shift your responsibilities to what you want.
Bio: Matthew Mayo is a computer science graduate student currently working on his thesis parallelizing machine learning algorithms. He is also a student of data mining, a data enthusiast, and an aspiring machine learning scientist.
Related: