Machine learning algorithms are can be supervised or unsupervised. Algorithms that are supervised apply what it learned earlier to new situations to predict future events. .The program can be trained to provide targets for new inputs based on what it has learned. It can also modify the model based on the error in the output. Algorithms for unsupervised machine learning use information that has not been classified or labeled. The program infers the structure hidden in unlabeled data. It explores the data and makes inferences from datasets to come out with information on hidden structures. The algorithm working on a mixture of labeled and unlabeled input data is called semi-supervised machine learning. The learning accuracy of such a program is fairly high irrespective of whether the issues of overfitting and underfitting exist. Semi-supervised learning is used when the labeled data requires relevant resources in conducting the training.
Another type of algorithm in use in machine learning is the reinforcement machine learning algorithm that discovers errors or rewards through interaction with its environment. This allows software agents to determine automatically the desirable behavior in a context to maximize performance. All the thousands of bots parsing website information and displaying advertisement on websites based on peoples’ browsing habits. They are basic forms of AI activity. The difference is that for supervised as well as unsupervised learning, there is constant mapping activity of input and output.