I’m looking for advice on how to implement a matching algorithm with some of the techniques learned in the class. An example of the type of setup I have is:
2 sets of data on cars with features like make, model, year, color, etc (not images). One set is a list of known possible cars with feature values that are known to be accurate. Another is a set of observed samples with feature values that may be inaccurate. I want to match the observed samples to the possible samples, with the catch that the observed car may not be in the known car list.
I’m thinking I need 2 models: 1 to find the closest match, and another to determine whether the observed sample and the closest match are in fact the same car. For the latter I can just feed in the sample features and the closest match features into a simple dense network ending in a sigmoid, but the former is stumping me. Is there a way I can use an autoencoder for this (like models that try to match a new image of a person to an on-record image of that person)?
Thanks for any advice.