Learn Machine Learning Roadmap - Prerequisites and Trend
New year new vision. Wish you a happy new year and a year of vision 2020. In today’s newsletter to all (subscribers and non-subscribers), we will share a best of the trend article (today we are sharing Google’s ML trend prediction for 2020), machine learning pre-requisites and also a few plugs why you should subscribe.
In this year we will release quite a few free articles, so stay tuned if you still want freebies but here are a few reasons why you should subscribe:
Why Subscribe
Only subscribers get easter eggs!
These are important resources sent directly into their inbox. For example Winter Quarter, subscribers received url links self driving car resources and Pytorch textbook giveaways.
January 2020 easter eggs are what is it like to be a machine learnist at work. Subscribers, your easter eggs will arrive soon! Stay tuned.
High quality articles. Subscriber exclusive articles.
The first email this year is a timeline of important discoveries and models, papers in machine learning and deep learning.
End-to-end machine learning course
Subscribers get access to end-to-end machine learning course. While this course is unofficial and still in development, subscribers get our best pointers on the nitty gritty part of machine learning. Since most of machine learning is spent on data cleaning and feature selection / engineering, Spring Quarter 2020 there will be a big release on Exploratory Data Analysis (EDA) and some of the lessons learned from Kaggle, Google owned Data Competition site.
Here’s a free article preview for the machine learning course
Free Resource for you today!
Google’s AI chief talks about important Machine Learning trends 2020. It is still a hot topic. We have ever reason to invest our time learning more.
Google AI chief Jeff Dean interview with Venture Beat: Machine Learning Trends in 2020
"At the Neural Information Processing Systems (NeurIPS) conference this week in Vancouver, Canada, machine learning took center stage as 13,000 researchers explored things like neuroscience, how to interpret neural network outputs, and how AI can help solve big real-world problems."
This important conference just happened. And as a newsletter, we take notes on behalf of our readers. Key takeaways:
BERT is everywhere
Specialized chip design in 2020 and optimization (in addition to model architecture improvement 2019) can improve ML performances!
Detailed notes will be a part of this month’s easter egg for subscribers.
Machine Learning Resources
We send the best articles we find in the internet to our subscribers. These articles are time savers. They include the best tutorials, resources, trend articles like the one you see above, courses that we love, timeline of the CNN models and their evolution, industry top reports.
Here’s a free preview of resources we sent out.
Friendly link to our best medium posts
Did you know that you can request “friendly links” to our Medium articles? We are popular on Medium and our article has been featured on KDnuggets and Medium! https://medium.com/@uniqtech See one that interests you on our publication? You can request a link for free via email hi@uniqtech.co
Machine Learning Prerequisites
Somehow we forgot to tell you what you need to know before getting started with Machine Learning. Gee.
While there are no real machine learning prerequisites but not knowing python can be very painful. That being said, I heard R is a nice data science programming language and functions like a nice API, and is easy to use. ggplot makes data visualization easy.
Machine learning is actually an interdisciplinary field that requires statistics, math, programming, AI, game theory, machine learning, deep learning and some computer processor, GPU knowledge… that being said, I think of 2020 machine learning more practically as a data engineering role rather than a data scientist role. The difference is data scientists can really benefit from being a PhD. They make decisions based on statistical knowledge. Data engineers can use the right tool to trial-and-error - why not just try all the models and see if something sticks on the wall. As long as you know how to code, and are a good engineer, chances are you can do well in ML. Did you know novices can win Kaggle data science competitions?
Pre-requisites
Working knowledge of coding in scripting languages like Python Ruby or JavaScript
Distance memory of Calculus and understanding of derivatives and integrals, chain rule. If you went through the class once, you will understand a whole lot more. You won’t be implementing these from scratch, there’s no need for a full review.
Nice to haves
Working knowledge of Object Orient Programming experience
Working knowledge of data science and data visualization techniques: experiences with SQL, Excel
Working knowledge of Cloud such as Google Cloud or Amazon Web Service (AWS)
Github code versioning experience
Really important stuff - learn asap
Python libraries that are great for data science: Pandas, Numpy, Matplotlib.pyplot, Seaborn.
Working knowledge of experiences using APIs (Scikit-learn, tensorflow, Pytorch are ultimately APIs for ML or Deep Learning)
Learn to read error messages: the errors tend to repeat, it will come with experiences. TypeError, SyntaxError, input errors params reading errors, dimension error is a big one when it comes to matrix operations
Care a little more about data. Remember ML models are like calculators, tools that you can use to operate on data. It is best to have interesting projects or goals before you learn ML. Check out Kaggle for interesting projects such as recognizing whales using computer vision, detecting skin cancer, visualizing the lung, clustering wines… and more.
Programming environment: the best to use is Google Colab, no installation nor configuration required. You can also use Anaconda here’s our guide on substack. Or if you prefer anaconda cheat sheet on medium.
Really nice to have but probably doesn't have
Working knowledge of linear algebra. You don’t have to review it. If you went through the class once, understanding deep learning will be a whole lot easier. Especially for image classification / computer vision tasks.
Game theory. Knowledge of game theory such as from Economics makes the robotics part of reinforcement learning a whole lot easier
Web development experience. Familiarity with Model-View-Controller framework. Some experience with Django / Flask (python MVC web development frameworks) for deploying models.
Data visualization techniques. Optional.
Statistics basics. Optional.
C++ is important for fast compute, fintech, self-driving, lower level deep learning customization, optimization.
Conclusion
It is a daunting list. But honestly if you find the right project the learning just flows! I used to not be able to finish data science books but now I write and mentor machine learnists, I fly through the pages because I am motivated by my inquisitive mentees.
Here’re two great examples:
Japanese farmer uses Tensorflow to intelligently categorize cucumbers for family farm
Japanese ramen enthusiast uses AutoML to intelligently categorize ramen from ramen shops!
Prefer a more formalized understanding of what’s required for you to get started on Machine Learning? I recommend the Udacity intro to machine learning nanodegree landing page and the syllabus on the page is excellent!