Efficient Quantum Machine Learning with Quantum Annealers: Overcoming Qubit and Connectivity Limitations for Classification and Regression Problems

Support-vector machines trained with quantum annealing can reproduce the performance of classical models, but current annealers restrict the size of the training set that can be embedded directly. This work addresses that limitation by combining quantum-trained support-vector machines with a local learning strategy that selects nearby training samples before optimization. The resulting approach supports binary and multiclass classification on larger real-world datasets and demonstrates how hybrid quantum-classical machine-learning workflows can work around present-day qubit-connectivity and problem-size constraints.