# What is the borderline between unsupervised learning and regular algorithms?

Unsupervised learning using neural networks is clearly machine learning since it is utilising neural nets.

However, some algorithms, k-means clustering, for example, are considered unsupervised learning, while they look just regular algorithms (non-ML).

What should be the borderline (criteria) to differentiate between unsupervised learning and a non-ML algorithm?

Any algorithm that uses data (in some form) to improve some performance measure (aka objective function), or to find some function, can be considered a machine learning algorithm. See this answer for more complete definitions of ML.

k-means does that. It uses the data to find some division of the data itself into groups, in order to maximize some objective function. So, k-means is a machine learning algorithm.

The use of neural networks is not a requirement for something to be called a machine learning approach. In fact, there are many other machine learning approaches/algorithms that do not use them, such as tabular Q-learning, support vector machines or hidden Markov models.

• tks, data+improvement is the point
– Dee
Aug 24, 2021 at 12:01
• data here are possibly statistical data to be specific, because some optimisation algorithms in graph theory are also optimising for better values after a number of iterations on the graph data
– Dee
Aug 24, 2021 at 12:53
• @datdinhquoc I don't know what definition of "statistical data" you're using, but by "data" I mean any data, including graphs. So, if that algorithm you're talking about uses data, such as graphs, and optimizes some objective function given more data, then it could be considered an ML algorithm too. In fact, there's an area in ML called "geometric deep learning", which is concerned with the application of deep learning-like ideas to graphs. If you provide the name of the specific algorithm, I may look at it later and tell you more what I think of it.
– nbro
Aug 24, 2021 at 14:58