Machine learning represents one of the dominant areas of artificial intelligence. This is why the term is sometimes mistakenly taken to be a synonym for artificial intelligence. However, artificial intelligence is a broader term: it covers other areas apart from machine learning.
Machine learning is an umbrella term for all methods and approaches that allow artificial systems to acquire knowledge, skills and to enhance or otherwise modify their behaviour in significant ways based on experience, which takes the form of data (whether pre-collected or gathered by the system on its own).
There are several kinds of machine learning – depending on the kind of task that the approach is designed to solve [Rojas1996,Chapelle2006]:
- Supervised learning: The system is provided with input-output samples. We know what the desired output of the system is for each input.
- Reinforcement learning: We do not know the exact desired outputs, but we tell whether the system behaves correctly and reward/punish it accordingly.
- Unsupervised learning: The dataset only contains inputs, there are no desired outputs and no rewards. The goal of these approaches can typically be described in terms of minimizing some cost function (examples include clustering, where the goal is to identify clusters in data).
- Semi-supervised: A hybrid kind of learning, where we only know the desired outputs for some of the samples and for others only the inputs are available. There are methods that can use even such unlabeled samples to improve the performance of the system.
Machine learning is closely related to the area of optimization: it typically involves some parametric function, the parameters of which are optimized w.r.t. some criterion.
Literature
- [Rojas1996] Rojas, R., 2013. Neural networks: a systematic introduction. Springer Science & Business Media.
- [Chapelle2006] Chapelle, O., Scholkopf, B. and Zien, A., 2009. Semi-supervised learning (chapelle, o. et al., eds.; 2006)[book reviews]. IEEE Transactions on Neural Networks, 20(3), pp.542-542.