Fundamentals of Machine Learning (simpler)
What (even) is Machine Learning ?
There is an official definition for it, but since this is all me breaking it down for you, machine learning is a method to make a computer do long calculations and understand large amounts of data, such that it can recognize all features and patterns present in it, and then make a decision as required. You have to specify what it has to learn, and what answer it has to provide.
What is machine learning based on ?
The functioning of machine learning, or artificial intelligence relies on mathematics, especially statistics and probability.
Machine learning algorithms use a variety of mathematical theorems as their basis, for eg. regression, Naive Bayes probability, 3-d vector geometry.
Fundamental terms used in Machine Learning
- Dataset : The dataset is the base for the machine learning model to be trained and tested on. It can be seen as a question bank with or without answers.
- Model : A machine learning model is an instance or ‘real-life implementation’ of a particular method or algorithm. It must be trained (and tested) to have it provide results.
- Training and testing : A ML model has to be trained on the dataset, and then be tested to verify its accuracy and performance.
- Training data / Testing data : The ML model is trained on the training data, and then tested, evaluated on the testing data. Both are usually obtained by dividing the original dataset.
- Accuracy : The accuracy of a model is usually a percentage of its accurate predictions against the total predictions.
What are the different basic types of machine learning ?
Supervised Machine Learning
In supervised machine learning , models are trained and tested using properly defined, labeled training data, and provide output in a specified format.
For example, in an MCQ test, a student is given questions as statements or expressions, and gives answers as options (a,b,c,d).
Similarly, in supervised learning, the model is trained on well defined data, with proper labels and values, and gives output in a specific format.
Unsupervised Machine Learning
Unsupervised machine learning is a bit trickier.
Suppose I give you a bunch of documents, fee receipts and report cards, and ask you to make sense of them.
Naturally, you would see if there are any of both that belong to the same student, or students of the same class, or printed on the same calendar date.
Essentially, you would try to find any relationship or pattern between the documents.
Unsupervised machine learning works similarly. It identifies relationships / patterns within unlabeled data and uses that to classify the data into separate clusters or derive any association.
Reinforcement Learning
Reinforcement learning mimics human behavior, as in, a model is allowed to operate (provide decisions/outcomes) in an environment (dataset).
Actions of the model that lead to the desired outcomes are encouraged and scored positively, while the ones that lead to undesired outcomes are discouraged and scored negatively.
The model, as a result, learns from its mistakes, and gravitates towards the actions/computations that are most desirable.
Popular branches of machine learning
- Deep learning : Deep learning mimics the human brain, using neural networks to provide processing and predictions on a large variety of data. It is a powerful concept and is in use in many domains.
- Natural Language Processing : Used to identify and efficiently process human language, as text, audio etc., and conduct a computation or decision making based on it, applying text mining, filtering, sentiment analysis etc.
- Computer Vision : Uses a computer’s capability to process and analyse images or video to extract meaningful information and use it to make decisions, applying object recognition,
