Machine Learning is the science (and art) of programming computers so they can learn from data.
Machine Learning is a subfield of Artificial Intelligence (AI). The aim of machine learning is to understand the structure of data and fit that data to model so that could be understood and utilized by people.
Think of Machine Learning as that super power in which organizations leverage on to decrease operational cost, increase return in investment and scale products etc by automation which involves collating data, training models and deploying this models to production to meet business or user needs.
Machine Learning is a field that comes under the umbrella of computer science, but it differs from traditional approaches of computation. In traditional computation, input and logic is provided in the form of code whereas in machine learning input, output and logic all are provided to generate model from current data and this model is used to predict output for the new data. It facilitates computers in building models from sample data to automate decision-making processes based on input data.
Types of Machine Learning Systems
Machine Learning systems has different types it is useful to classify them in broad categories based on:
- Whether or not they are trained with human supervision (Supervised , Unsupervised, Semi-Supervised and Reinforcement learning)
- Whether or not they can learn incrementally on the fly( Online versus Batch learning)
- Whether they can work by comparing new data points to the known data points or they try to detect patterns in training data and build predictive models ( Instance based learning or model-based learning)
In supervised learning the input data you feed to the model also contains the target labels. Model tries to learn from input labels and target labels and after learning models make predictions. Supervised learning deals with two typical tasks i.e classification and regression. A spam filter is a good example of classification task, it is trained with many spam emails or ham emails and model learns from data how to classify new emails. Predicting housing prices is an example of regression, input provided data contains features of house like size, room numbers etc with target label price and model predicts the continuous value.
These are the most Common Supervised Algorithms:
- Linear Regression
- Logistic Regression
- K-Nearest Neighbors (KNN)
- Decision Tree and Random Forests
- Naive Bayes
- Support Vector Machine (SVM)
- Neural Networks
In unsupervised learning input data is provided to system without target labels and model tries to learn from data without labels. The aim of unsupervised algorithms is to model the underlying structure or distribution of data in order to learn more about data. These are called as unsupervised learning because far from supervised learning the labels are not available in data model tries to learn without teacher by finding out the interesting patterns from data. Identifying drug molecules from the set of small molecules is an example of unsupervised machine learning. The molecules can separated by evaluating the similar pattern in their drug-like properties and thereby two different sets of molecules will be obtained.
These are the most Common Unsupervised Algorithms
Clustering problem is where your algorithm discover the inherent grouping in data, these groupings are based on the similarities distance.
- K-Means clustering
- Apriori algorithm for association rule learning
- Principal component analysis
In Semi-supervised data you provide to your model contains a little labeled and a lot of unlabeled data. Semi-supervised problems resides between supervised and unsupervised learning. Many real world machine learning problems falls into semi-supervised learning area, because it is very expensive to acquire labelled data and unlabeled are cheap to acquire. Semi-supervised learning uses both unsupervised and supervised learning algorithms. By using unsupervised technique it is easy to discover the patterns and learn the structure of the input variables. These patterns of data can be feed to supervised learning algorithm as training data and use the model to make predictions on new unseen data.
Photo hosting services such as Google Photos are good example of this. When you upload all your family photos, it automatically recognize the person A shown up in 2, 4 and 9, while another person B shows up in photos 3, 6 and 8. That is unsupervised technique of clustering. You only need to label one person and system will able to label other photos.
Reinforcement learning is different beast, where instead of feeding inputs, system learn by itself on the basis of rewards and penalties for every decision making. The advantage of reinforcement learning is its dynamicity as learning develops with changing environment. In this type of machine learning, learning system is called an agent. Agent is able to observe environment, select and perform actions and get rewards in return or penalties in the form of negative rewards. Agent must be able to learn by itself and based on learning should need to define policies. Policy defines what action agent should need to pick in a given situation.
Robots are designed on the basis of reinforcement learning algorithms to learn how to walk. DeepMind’s Alpha GO program is also a good example of reinforcement learning, it has beaten the world champion Lee Sedol at Go game. It analyzed the wining policy of millions of games and playing many games itself.
Machine learning is the subfield of artificial intelligence. Instead of relying on explicit programming, it is a system which uses massive data sets and apply algorithms to train and to teach themselves to make predictions. Those problems which requires lot of time to solve using traditional approaches machine learning can solve those problems with in short time. Machine learning is categorized into four major types supervised, unsupervised, semi-supervised and reinforcement learning.
Classification of diabetes using latest Automl framework FLAML