Accelerating Neural Network Ensemble Learning Using Optimization and Quantum Annealing Techniques
Ali Jooya, Babak Keshavarz, N.J. Dimopoulos, Jaspreet S. Oberoi · 2017
Neural networks (NN) are models of choice in machine learning, as they can learn and model highly non-linear functions. However, the process of training NNs is an ill-posed problem, as the solution is either not unique or not a continuous function of the input data. We have previously introduced an ensemble NN model that is less ill-posed by employing a) heuristics in the training phase to develop models with good generalization capability and b) a post-training Sensitivity Heuristic to exclude the models that are sensitive to input perturbation. Although the heuristics have improved the accuracy of the model, they are time consuming and perhaps non-optimal.