We look briefly at the fundamentals of supervised learning, particularly in the context where observations are not noisy. We note the importance of feature sets that have Universality, mentioning various feature selection and hierarchical methods.
We then give several examples where the basic concepts of machine learning can add value in finance – for example in determining the dimension of the investment space.
Finally, we observe that sensible feature sets for functions on the space of unparameterized paths or streams can be built, parsimoniously, from the signature (and group structure).
Using these features, and the mapping of rich data to continuous date using the Hoff process, we give some simple applications using real financial data.
Back to Broad Perspectives and New Directions in Financial Mathematics