Machine learning "teaches" computers to find solutions from data. Its quality gets better whenever there is more, or higher quality data given to it.
There are three families of machine learning:
- Supervised learning
This family of models trains on labeled data, meaning the dataset included the correct output. These moduls include classification models, regression models, and support vector machines. There is also deep learning, which analyzes data through a layered structure of algorithms inspired by the neural network of the human brain, leading to more effective learning. - Unsupervised learning
This family trains on unlabeled data, which means that the algorithm needs to discover patterns and structures without any human guidance. Two main types of unsupervised learning are dimensionality reduction, which consists in mapping a multidimensional dataset into more interpretable two-dimensional structures, and clustering, which groups observations into classes based on shared characteristics. - Reinforcement learning
This family relies on receiving feedback via rewards or penalties. It used mapping between input and output like supervised learning. It includes a feedback loop that gives the algorithm positive and negative signals, so that it adjusts accordingly. This type of training needs massive amounts of data, long computation time and there is limited generalizability.