DEV Community

dinhanhx
dinhanhx

Posted on

How did I start with Machine Learning?

As of writing (2021), I am last-year ICT Bachelor student of USTH.

2019

When my second year started, I knew nothing about machine learning nor artificial intelligent. I just knew these words existed. During this year, I learned Probability and Statistic, Digital Signal Processing, Image Processing, Numerical Method. My GPA for these subjects are not significant. I admitted that I am not good at math not I am still not. If you tell me to multiply 3 by 3 matrix with 3 by 2 matrix, I will probably do it badly. Fortunately, I'm not so bad at Calculus.

2020

The covid outbreak in Vietnam, I had more time for myself and staying at home. I watched a lot of videos on You Tube about people making AI to do this do that. Code Bullet is one of my favorite channels. There was a professor who was working on forecasting political trends from news. He sent an invitation to my class. I signed up then I told him I didn't know anything about machine learning. He said that was fine then let me join the project. Most of the time, I did not understand any concepts or any terms. He explained what he was trying to do. My job was to label news manually. In order to learn basic concepts, I googled a lot. I found two ways of machine learning. There are top-2-bot and bot-2-top.

  • top-2-bot: you learn what is AI, ML; what are the problems; what are the methods, till you reach math

  • bot-2-top: you learn math, learn programming, learn applying the math into the programming, learn the methods, know the problem

I didn't follow one of them. I followed a way that I felt easy. I was not good at math, knew nothing about ML. As a result, I should choose top-2-bot, right? I was too dumb to run a neural network with instructions. The way is to combine both ways.

To know what is what, I use top-2-bot to learn general idea about machine learning. The blog I read is Start here with Machine Learning. Then I read Machine Learning with Python.

To learn how does it work, I use bot-2-top. You should apply this technique with small algorithms such as linear regression, logistic regression, etc.

To learn how to apply the method, I use top-2-bot because practice make perfect. You may treat a method as black box. You understand the input, the output and the idea of it. You should apply this technique with neural network or probabilistic models. For neural network, I read Deep learning with Python.

2021

From a guy who is bad at math (still) to the guy know to apply Mask RCNN to segment nodule from lung ct scan is a long journey. I feel more confident and am ready to learn more.

Top comments (0)

Sentry image

See why 4M developers consider Sentry, “not bad.”

Fixing code doesn’t have to be the worst part of your day. Learn how Sentry can help.

Learn more