My first exposure to the potential of quantum computing came at a machine learning conference. Naturally, this talk resonates with me even if the title is a mouthful.
This talk was presented live by Pinaki Sen, Engineering Undergrad at NIT Agartala, India, and Research Intern at ISI Kolkata, India.
Recently, the closeness between the techniques used in classical machine learning and quantum-many body physics has got significant attention among the academic and research communities. Especially, the deep learning frameworks and tensor networks hold similar properties such that they can be used for machine learning tasks. In noisy intermediate-scale quantum (NISQ) technology, the quantum circuits with a long circuit depth or a large number of qubits cannot be implemented on NISQ devices. It is highly demanding to develop applications with adequate resources which can exploit the quantum advantages.
In this paper, we proposed the architecture of hierarchal extreme quantum machine learning consisting of quantum tensor variants of autoencoder (i.e. Tree tensor network (TTN), multi-scale entanglement renormalization ansatz (MERA)) and quantum neural network (QNN) for binary classification. The proposed hierarchical architecture has the ability to overcome the shortcomings of regular tensor networks and can be defined in complex geometries efficiently as long as the order can be represented appropriately.
We apply the quantum variants of TTN and MERA for image compression and quantum neural networks to classifying the images and compare their performance, concluding that the combination of TTN performs better than MERA with QNN for image classification.