There may be quantum methods which could be used to improve current gradient descent
and other learning algorithms. It may also be possible to combine some of these with a
quantum search. An example would be to use gradient descent to try and refine a composite
weight vector found by quantum search. Conversely, a quantum search could start with
the weight vector of a gradient descent search. This would allow the search to start with an
accurate weight vector and search locally for weight vectors which improve overall performance.
Finally the two methods could be used simultaneously to try and take advantage of
the benefits of each technique.
Other types of QNNs may be able to use a quantum search as well since the algorithm
only requires a weight space which can be searched in superposition. In addition, more
traditional gradient descent techniques might benefit from a quantum speed-up themselves.