WHAT IS MACHINE LEARNING AND WHAT ARE ITS TYPES?
Today machine learning has become one of the main technologies used in market, but why ? what is machine learning? Why are companies really spending billions of dollars in AI AND MACHINE LEARNING.
the answer to above question is very simple, the only reason is- machine learning starts where naïve computing fails and is too complex .
Here is the example for a problem that is too complex for naïve computation, we all use YouTube and we often see videos and shorts in it, so how does we get recommended videos ? Well for this ,we can use an algorithm (not ML) and make a program that collects videos and group them using label .We can do a sort of recommendation system.
Hmmm , you might think that this solution is perfect, but it has some problems, most important is that every time new trend (like a new field of cs has emerged) has come , we have to separately make groups and also if some if trends or groups becomes part of another group then we again have to seperately make another alogrithm for merging some groups or it may be possible that a trend has gone (lets someone decide that after some years all videos related to corona should be deleted as no one wants to remind of so called dark days) so we have to make another program for deletion.
Due to all this problems , we use machine learning . But how does this solve above problem? Answers to these question will be given when we will make a real recommendation project. So let’s start with Machine Learning.
What machine learning theoretically it is-
“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E”.
Naïve definition — It is a program capable of learning by itself and also adapting to new conditions without writing extra codes. This might seam pretty good.
Where we can apply machine learning- where problems are too complex or where we don’t have any specific algorithm for solving that problem.
MACHINE LEARNING TYPES-
SUPERVISED LEARNING-
Here we feed data along with labels.
It simply means that when we give data we also provide desired solution for it. So lets take an example to find out what exactly it means, when make a model that classifies an image as train or not, but how does model knows what is train and how does it look.
That is why we give input data as data with labels ( that is 1 if image is train or 0 if not). This is known as supervised learning. Diff models in supervised learning are Linear Reg. , Logistic Reg., SVM, KNN, Decision Tress.
Supervised learning are mainly used for prediction or using as binary classifiers.
UNSUPERVISED LEARNING
Here we feed data containing no labels.
We give only input data and algorithm by its own identifies different clusters and when new data comes it successfully identifies in which clusters it belongs to.
We use unsupervised learning for classification or clustering , dimensionality reduction ,anomaly detection ,etc. Some important models in it are K means , isolation forest , kernel PCA , etc.) .
REINFORCEMENT LEARNING
Here we say that model learns from experiences and get rewards and punishment depending whether desired output has obtained or not.
here agent is learning system, model must learn by itself and develop a best strategy called policy to get reward most of time and lastly policy defines what actions agent should choose in a given situation, example of reinforcement learning is self driving cars (many might be thinking of TESLA) .
BATCH LEARNING v/s ONLINE LEARNING
Batch learning- here system doe snot learns as we feed data, simply model is trained and is launched and runs without learning. It is also called offline learning. Major problem of this is that if our data is changed or we have to train a new data then we have to again train our model from scratch.
Online learning- as the name suggest system learns from data we feed. Here data can learn about new data and model can adapt to change very easily .
INSTANCE BASED LEARNING v/s MODEL BASED LEARNING
Instance based learning- system learns training examples by heart , when new data comes , it generalizes it according to the learned data or by using a similarity measures
Model based learning- we build models and use that nodels to make predictions
+++++++++++++++FOLLOW ME++++++++++++++++++
