What is Machine Learning?
Nowdays the term ‘machine learning’ is used as an magic word. Everybody talks about machine learning like it solves almost any problems we have. But if you look into it, you will know it is just a chunk of fancy mathematical programming codes.
Why named it machine learning:
What we generally think of AI is AGI, artificial general intelligence, which we can see in lots of SF movies. But AGI is very difficult to make as we don’t understand what is counciousness.
Whenever we refer to AI, it usually means machine learning or deep learning model. The models are just a chunk of super fancy mathematical programming codes.
Just think of machine learning as making some kind of function. All you have to do is just input some data and get a result. If input is a dog or cat picture, the result is to tell what is dog and what is cat.
So machine learning is very simple concept except that the function is generated by so many mathematical calculations. And that’s why we call it ‘learning’.
Machine learning vs Deep learning:
Machine learning is a subset of artificial intelligence(AI). And you might have heard about deep learning also. Deep learning is a subset of machine learning.
What’s the difference between machine learning and deep learning? It is easy to tell. Deep learning model includes feature extraction, while machine learning model does NOT include it. Machine learning is broader concept than deep learning.
Feature extraction? When you describe a car, what is the fetures of it? A car has doors, a wind sheid, wheels and so on. These are features of a car, and you just ‘extracted’ the concept of features from it. This is feature extraction.
Data preprocessing:
Keep in mind though, you can’t just input any data you want in machine learning. The raw data should be preprocessed to make a model with good performance. Data preprocessing includes the followings:
- Data Cleaning : It’s the process you have to clean the data set, as there are some missing information or irrelevant data. Fill in missing ones or remove unnecessary or redundant ones. You need high quality data sets.
- Data transformation : There are two types of data. Structured data and unstructured data. structured ones are text file, csv file or json file. Unstructured data includes images, sounds etc. You need to make unstructured data set structured by encoding.
- Data validation : In this stage, you divide data into several sets, especially test sets and training sets. The training data is for training ML models. The test data is for testing whether the model is trained well or not.
MLOps:
You are now ready to make a machine learning model with bunch of refined data sets. but just training and making a model is not enough. The model needs to be cared continuously. It’s the MLOps step.
MLOps means machine learning operations. It is series of steps to run AI model successfully in real business. Its whole process looks similar to that of DevOps. What makes MLOps different from DevOps is that it includes data science skills.
