View profile

Machine Learning in Production

Machine Learning in Production
By Santiago • Issue #22 • View online
I’m currently going through the “Introduction to Machine Learning In Production” specialization. I’m on the second course already.
A little bit every day. Slow, and taking my time with every lab.
I’m not just checking the boxes but making sure I go down deep into as many rabbit holes as I can. It’s mind-blowing how much you can learn when focusing all of your attention on a good course.
The specialization is all about Machine Learning in the industry. It is as practical as it gets.
I can’t recommend it enough.

Contrastive Learning
This is a cool example.
It tries to solve the MNIST digit-classification problem using Contrastive Learning. A different and interesting way of solving a very popular problem.
You’ll find the code below. I used Deepnote. I’ve been very impressed with it, and I’ve started to use it daily.
If you are looking for inspiration on how to apply interesting techniques to boring problems, start here.
Classifying handwriting digits using the MNIST dataset is the most popular computer vision problem out there.

Everyone has solved this problem the same way:

• Using a fully connected network
• Using a convolutional neural network

That's boring. Let's do something different.
Fake data is fake
One of the most common problems we face: not enough data to build a good model.
The obvious solution is to generate synthetic (fake) data. This is good, but it comes with tradeoffs.
On this thread, I’m trying to argue that fake data is a great way to augment a dataset, but most of the time, it shouldn’t make for the majority of the data. The keyword here is “most of the time.”
The first recommendation I always hear whenever I start asking about data:

"We can create fake data!"

Unfortunately, it's not that simple.

Thread: On synthetic data and when and how to use it.
In pursue of a Machine Learning career
I’ve posted this before. It was very popular, so I decided to post it again.
Here are 5 courses/specializations that you can get from Coursera. They are long. They are hard. They are worth it.
I’d recommend you take them in order.
5 Coursera courses to become a Machine Learning Engineer:

1. Machine Learning
2. Deep Learning Specialization
3. TensorFlow Developer Professional Certificate
4. TensorFlow: Advanced Techniques
5. Introduction to Machine Learning In Production

Take them in that order.
Did you enjoy this issue?

I'll send you an email whenever I publish something important. I promise you don't want to miss this.

In order to unsubscribe, click here.
If you were forwarded this newsletter and you like it, you can subscribe here.
Powered by Revue