Types of Machine Learning : New Approach with Differences

You guys are mostly familiar with the Trending word Machine Learning . Some of you also know the types of Machine Learning . So you must be wondering what value you will get in the article . See , We all know generally , There are 3 types of Machine Learning : Supervised , Unsupervised , reinforcement Learning . Some of us have also read about semi supervised learning as hybrid of supervised and unsupervised learning . You all are right buts its just a classification based on some set of parameters . There are few more sets of parameter which we use to classify machine learning . As a machine learning engineer , data scientist or AI engineer , I think you should know these basics . So article will open new cards on Types of Machine Learning : New Approach with Differences .

Types of Machine Learning –

There are three way to classify machine learning –

Batch Machine learning Vs Online Machine Learning .
Instance based Machine Learning and Model based Machine Learning .
Supervised vs Unsupervised vs reinforcement Machine Learning .

Batch Machine learning Vs Online Machine Learning –

Batch Machine Learning is more popular with the name of offline Learning . This is not real time incremental learning . You need to train the Model with all available data at once . In case you train and deploy the model , Now you have produced some more training data . You need to mix that new generated data with older one and re train a new model . Now you may replace the old model with newer one . This retraining may require lot of training time and computing resources .But It ensure more hold on accuracy .

In the opposite side Online Machine Learning is a type of Machine Learning where we keep on the training the current model real time with new generated data . This involves less computation but there may garbage data enter into system and pull back the accuracy as well . there is a another name for this technique is out-of-core learning .

In General when we talk out Machine Learning , We focus on Batch Machine Learning . Hence most of the implementation ( code ) in popular machine learning algorithm frameworks contains the interface for Batch Machine Learning . Still Scikit-Learn Itself provide the interface for Online Machine Learning for below algorithms.-

Feature extraction: Mini-batch dictionary learning, Incremental PCA .
Regression: SGD Regressor, Passive Aggressive regressor.

There are few more examples for but the motive here is to introduce you with the terminology only .

Instance based Machine Learning and Model based Machine Learning –

Instance based Machine Learning finds the similarity of new observation with existing data . It suggest the same scenario of past with highest similarity score with current data . It is easy for you to relate with some real examples . RBF networks , K-nearest neighbors algorithm are quite general and popular example for Instance based Machine Learning .

Model based Machine Learning is generalize kind of machine learning . Where we build the model using existing past data . Once we have the model , We predict using the machine learning model . Examples are decision tree , random forest , Logistic regression etc .

Supervised vs Unsupervised vs reinforcement Machine Learning –

In general When we categories Machine Learning , We get the name Supervised , Unsupervised and reinforcement learning . Supervised Machine Learning is a technique where training data contains the desire variable (target variable ) . In other words this learning has labeled training data . There are so many examples for supervised machine learning algorithms .Linear Regression ,Logistic Regression ,k-Nearest Neighbors ,Support Vector Machines ,Decision Trees and Random Forests and Neural networks are good examples of supervised machine learning algorithms .

Unsupervised Machine Learning is a technique where training data is not labeled . Actually Training data do not contains the outcome variable . k-Means , Expectation Maximization ,Hierarchical Cluster Analysis , Apriori , Eclat are good examples of Unsupervised Machine Learning .

Reinforcement Machine Learning is quite different than above two techniques . Here the learning system ( agent ) take some action and gives the reward for it . Reward can be negative or penalty for a wrong action . The agent keeps train it self and the generate its policy by rewarding . In the real world AI has quite evolving in this stream . Q-Learning , Asynchronous Actor-Critic Agents (A3C) , Temporal Difference (TD) , Monte-Carlo Tree Search (MCTS) are best example for reinforcement learning .

Now a Days a new kind of machine learning is also popular semi supervised learning . It a hybrid technique of Supervised and unsupervised machine learning .

Thanks

Data Science Learner Team