For the problems below, choose which technique would provide the best approach: a supervised neural network, an unsupervised technique or non-linear regression with a known functional form.

You have lots of examples of chemical spectra, and you know what each one is. You want an algorithm that will automatically classify new spectra into one of the known categories.

You have lots of astronomical observations of galaxies, and you want to know if there are different types of galaxy (and how many different types there are). You know nothing about the possible categories.

A bank has a variety of data on customers, including their age, income, address, etc, and whether they defaulted on their mortgage. They want to develop a model for predicting future mortgage defaults.

You have a quantity that is known to obey an approximate polynomial relationship. You want a model that predicts that quantity for new data points.

You want to identify new types of insurance fraud in a large data set of insurance claims.

A particle detector at the Large Hadron Collider records the responses of different particles as they scatter from a collision event. You can simulate the behaviour of each particle type in the detector, and want to develop a model to classify the particle type that is observed in real event data.