There can be various ways of Grouping Machine Learning Algorithms. The most common way is to organize the Algorithms based on Learning Style and the following are the categories:
1. Supervised Learning
2. Unsupervised Learning
3. Semi-Supervised Learning
4. Reinforcement Learning
Supervised Learning:
In Supervised Learning, we, the human experts, act as the teacher. We are feeding the machine(computer) with some training data which contains a given set of predictors(i.e. independent variables) and an expected outcome (i.e. dependent variable). Using these sets of variables, we generate a function that maps the inputs to the expected outputs. The training process continues until the model reaches the desired level of accuracy on the training data.
Then the computer should be able to learn the patterns from these training data and then predict some test data.
Now, how I see this is like teaching my toddler about something. For example, I took my kid to a Zoo. I showed her a Zebra and told her that a Black-n-white animal is a Zebra. Then we saw a Black-n-White Penguin and she excitedly screamed “A Zebra”! So, I told her ‘No, that is a Penguin and it has a funny walk on 2 legs’. So, she registered that 4 legs, the black-n-white animal is Zebra and 2 legs Black-n-white animal is a Penguin. Now, what if we see a Panda?
So, here is my example of Supervised Learning and the DataSet is absolutely my own creation(Please pardon me if any information is incorrect in the below Data):
|
Independent Variables(Predictors)
|
Dependent Variable(Outcome)
|
||||
|
Sr.No
|
Color
|
Group
|
Eating Habit
|
Habitat
|
Name
|
|
1
|
Black-and-White
|
Horse Family
|
Grasses
|
Africa Grasslands
|
Zebra
|
|
2
|
Black-and-White
|
Birds
Family
|
Fish
|
Antarctica
|
Penguin
|
|
3
|
Black-and-White
|
Bear Family
|
Bamboo
|
China
|
Panda
|
|
4
|
Black-and-White
|
Marine Mammals Family
|
Fish
|
Antarctica
|
Killer Whale
|
In the above example ‘Color’, ‘Group’, ‘Eating Habit’ and ‘Habitat’ are the predictors and the name of the animal is the Outcome. So, with the above DataSet, we train a machine to predict an animal name based on the above features.
Now, if we introduce a new Black and white animal that belongs to Horse Family and eats Grasses and belongs to Africa, the Machine will be able to predict that is a ‘Zebra’.
- the algorithm relies upon an assumption made by humans about how the computer should learn and these assumptions can bring in a bias.
- Also, the human assumptions are based on the Training Dataset which may be incomplete or not sufficient for a prediction at Global Level.
Below is the list of common algorithms for supervised learning:
- Linear Regression
- Logistic Regression
- Polynomial Regression
- Decision Trees
- Random Forest
- K-Nearest Neighbors
- Support Vector Machine(SVM)
Unsupervised Learning:
In Unsupervised Learning, the machine is feed with some Unlabeled Data which helps in deriving some meaningful insights on the data. Here, we do not have any Outcome Variable to predict. This is the family of machine learning algorithms which uses some techniques on the input data to detect some patterns and then group the data based on some independent variables. These are mainly used in pattern detection and descriptive modeling.
In my above example for animals, suppose the outcome variable i.e. name of the animal is not there. We did not train the machine about these animals and thus the machine has no idea about those. So, the computer will try to look for some patterns. For Example, it may divide the four animals into 2 groups Herbivorous and Carnivorous. Or, It may arrange the animals into 3 groups based on Habitat.
Below is the list of common algorithms for unsupervised learning:
- K-Means Clustering,
- Association Rules
Semi-Supervised Learning:
As the name suggests, Semi-Supervised Learning is halfway between supervised and unsupervised learning. It means the Learning where typically a small amount of the Data is Labeled and a large amount of data is not. Some machine-learning researchers believe that unlabeled data, when used in association with a small amount of labeled data, can provide a significant improvement in learning accuracy.
The common methods for Semi-Supervised Learning are as below:
- Generative models
- Low-density separation
- Graph-based methods
Real-Life Example of Semi-Supervised Learning:
1. Google Expander
2. Speech Analysis
3. Web Content Classification
Reinforcement Learning:
Reinforcement Learning is the type of Machine Learning where it allows the machine to interact with the environment in an iterative way and gather some observations and then take an action which is of minimum Risk and Maximum Reward. This model of learning by trial-and-error, entirely from rewards or punishments, is known as reinforcement learning. The machine learns from its past experience and tries to capture the best possible knowledge to make accurate business decisions.
List of Common Algorithms are as below:
- Q-Learning
- Temporal Difference (TD)
- Deep Adversarial Networks
Real-Life Example of Reinforcement Learning:
1. Computer played board games like Chess, Solitaire, Backgammon
2. Robotics in manufacturing industries to pick a device from one box and putting in another
3. Self-driving cars.
Thank You!