In the last blog, we have seen the Machine Learning System based on Human Supervision(Please read it here). In this blog, I will continue the discussion of Different Machine Learning Systems. Here we are going to learn about Systems based on incremental learning ability and Prediction and Detection.
After reading this please leave your comments and reviews in the comment box down below, that really helps me to be motivated and improve my blog, so that everyone who is reading this will get the best out of it.
"*Data will talk to you if you're willing to listen."
—— Jim Bergeson*
Machine Learning Systems based on Incremental Learning Ability
Incremental Learning is learning while running the system. ( You will understand it below ) When we create a Machine Learning System, there are two different approaches to pass the data to the Model. Based on these two approaches we can divide Machine Learning Systems into two different types as follows:
- Batch Learning
- Online Learning
Batch Learning
In this Machine learning System, we pass a batch of data only at the beginning stage when the model is being trained. After the model is ready and is launched then it cannot accept new data to learn more. Here data is passed at only the training stage. Due to this, the system takes a long time for getting ready for use.
Also, if your data is huge, training all of your data at once will require a lot of computing power at it can cause a lot of money as well. Sometimes your data can be so huge that you won't be able to use Batch Learning Process.
If you want to pass new data to your existing model, you have to re-train it on the new data( means new data plus old data) and then you can replace the new learned model with the old model.
Online Learning
Here, Initially, the system is trained on small data and becomes ready to run. And even after launch the system can accept the new data as individual instances or in small batches and train itself while running. Due to this, the system becomes less time and resource-consuming. Because of small instances of data, it takes less time to get trained.
The big Challenge for Online Learning is that if false data is passed to the running model, the system can experience a drop in its performance. Suppose, some sensors of your robot got into the problem and it's sending false data to the System then it can create a problem. So to avoid this type of problem you need to closely monitor the performance of the system.
Machine Learning Systems based on Prediction or Detection
We can also categories the Machine Learning systems based on different use cases like Prediction or Detection. Let's study those below.
Instance-Based Machine Learning System
Instance-based system means here we will be considering a single instance and we will be detecting the group of data that is similar to this instance using some similarity measures.
Consider a Spam email filter example, here to flag the given email as spam we have to compare it with other emails, and using some similarity measures we will find the best group of data. Here we are finding a similar pattern of data. We are detecting the pattern in data.
Model-Based Machine Learning System
Here what we do is, with the help of training data we will create a mathematical model that will be used to predict the desired result.
For example, consider data of features of a house and its price where features of a house show a linear relationship with the house price. Then we can create the linear model and use it to predict the house prices for other houses using their feature and our created model.
Conclusion
So, here you learned about the different types of machine learning systems. You can choose the best Machine Learning System according to your particular use case.