Multi-class quantum classifiers with tensor network circuits for quantum phase recognition

Hybrid quantum-classical algorithms based on variational circuits are a promising approach to quantum machine learning problems for near-term devices, but the selection of ansatz is an open issue. Recently, tensor network-inspired have been proposed as natural choice such ansatz. Their employment binary classification tasks provided encouraging results. However, their effectiveness more difficult still unknown. Here, we present numerical experiments multi-class classifiers tree network and multiscale entanglement renormalization circuits. We conducted image with MNIST dataset phase recognition XXZ model by Cirq TensorFlow Quantum. In former case, reduced number classes four match aimed output 2 qubits. The data consist three ground states prepared checkerboard circuit used eigensolver, corresponding distinct phases. Test accuracy turned out be 59%-93% 82%-96% respectively, depending architecture type preprocessing.