Types of Machine Learning
Supervised learning is a very basic form of Machine Learning where the human intervenor (you, the programmer) has a set of labeled data, which then is fed to the neural net.The neural net is trained iteratively as it goes through the data, with immediate feedback, due to the labeled data providing ground truth. For example: if our data set is pictures of cats and dogs, labeled as so, the neural net also knows this as it goes through and trains.
Unsupervised learning is similar in that we have an established set of data, but it’s not labeled. The neural net goes through the data and establishes patterns, although it may not understand the meaning of the patterns. There is no predefined ground truth in unsupervised learning. Instead, the model is tasked with things like finding relationships, anomaly detection, clustering (grouping similar things together, ish), or dimensionality reduction (downscaling while maintaining essential characteristics, kinda).
Transfer learning is kind of the idea behind “fine tuned” models. The idea here is taking one pre-trained model and using it as the base for another model.
Reinforcement learning is kind of like supervised learning. The model receives positive or negative feedback each iteration, and adjusts its weights based on that. A good example of this would be a model that plays Go — it has a clear reward or punishment each iteration. It can adjust its weight based on if it wins or loses.