Abstract by Seth Donaldson
Metalearning Using Metadata
A fundamental theorem of machine learning states, in essence, that there does not exist one single learning algorithm that can solve every conceivable problem with a high degree of accuracy. I.e., there is no universal algorithm. To solve this problem, we propose that machine learning can be employed to select the best possible algorithm (or set of algorithms) for a given dataset and problem.
Our research to date has been focused on generating metadata about a dataset that can be learned from. My work in particular has been focused on verifying the validity of our metadata generation by comparing our generated metadata against the metadata generated by OpenML (a nonprofit organization dedicated to data-science) on the same dataset.