Quantum Machine Learning Stephan Barabasi Abstract n n
Quantum Machine Learning Stephan Barabasi
Abstract n n n This paper explores the use of quantum computing for solving machine learning problems more efficiently and/or complement classical methods. Today only relatively small machine learning problems can be solved using quantum machine learning methods due to the fact that present quantum computers are rather primitive, and quality of results is affected by noise and decoherence. Nevertheless, the promise of future benefits that will substantially outweigh classical machine learning, make quantum machine learning attractive to data scientists and researchers. Pace University Students learned and practiced Quantum Machine Learning part of CS 837 class Pace University Students performed a series of experiments and published projects using Quantum Machine Learning
How Quantum Machine Learning started? n n QML started with the work of Francesco Tacchino and team at University of Pavia in Italy, implementing the world’s first perceptron on a quantum computer. The perceptron is a single-layer neural network, defined originally by Frank Rosenblatt and presented in 1958 at the US Office of Naval Research. Tacchino used IBM’s five Qubit “Tenerife” Superconducting Quantum Computer to take a classical vector, like the bitmap of an image and combined it with a quantum weighting vector A quantum computer can process exponentially more dimensions vs. classical systems. Tacchino’s experiment processed 2 N dimensions on a quantum perceptron vs. 1 x. N on equivalent classical computer IBM Tenerife (ibmqx 4) qubits connectivity
The evolution of Quantum Machine Learning (QML) n Initially, quantum computation helped implement complex feature map as quantum states into quantum computer provided hyperplane and subsequently enabling scalability of kernel methods computation using quantum computation. n n n Such solutions use classical data and quantum computation only augmented the classical algorithm implementing machine learning. Examples for this are QSVM (Quantum Support Vector Machine) and various classification algorithms Wikipedia identified three approaches Quantum Computing provides contribution to machine learning: n n n Only quantum data Only quantum algorithm Both quantum data and algorithms at the same time QML applicability approaches
How to image QML in the space of Artifical Intelligence? n n The Venn Diagram to represent best the various context for Artificial Intelligence and Machine Learning, Deep Learning and QML The evolution of hardware expanded the horizons of QML due to better computational capabilities that became available. QML
Quantum Machine Learning n n n Quantum Machine Learning(QML) incorporates machine learning models that can leverage quantum properties. The first QML applications focused on refactoring traditional machine learning models so they were able to perform fast linear algebra on a state space that grows exponentially with the number of qubits. The evolution of quantum hardware have expanded the horizons of QML evolving onto heuristic methods which can be studied empirically due to the increased computational capability of quantum hardware. AI QML became the next layer within Generic Machine Learning projects areas. QML has two main components: n n n a) Quantum Datasets b) Hybrid Quantum Models
Technology and project areas evaluated by Pace Quantum Computing Class (CS 837/CS 737) n Xanadu. AI n Photonic Quantum Machines n Pennylane Platform n n Unlike electrons, photons are very stable and are almost unaffected by random noise from heat. Integrates IBM, Rigetti, Google SDKs Tensorflow Simulator used via Google Colab IBM: n Superconducting Machines n IBM QASM Simulator part of Qiskit SDK Rigetti’s: n Superconducting machine n QPU: quantum chip with multiple qubits n QVM simulator: based on Py. Quill SDK Coldest place in universe: ~ 0. 015 Kelvin
Contributors to QML (1 of 3): IBM and MIT https: //ibm-q 4 ai. mybluemix. net/ We evaluated and experimented with: n QSVM for Classification n Quantum Image Recognition: n n Recognize hand-written digits Experimentation for license plates recognition
Contributors to QML (2 of 3): Xanadu. AI Algorithms and solutions (tutorials) evaluated: n AI Accelerators n Variational Quantum Linear Solver (VQLS) n n Quantum Classifiers using n Quantum Variational Classifier n n n Solve problems efficiently with multiple linear equations Using quantum computers as AI Accelerators Classical to Quantum Transfer Learning Quantum Variational Eigensolver (VQE) Neural networks type applications n Quantum Convolution Using quantum devices as neural networks
Contributors to QML (3 of 3): Google Tensorflow Quantum Google’s Cirq SDK: n n n Implements Quantum Operations Cirq defines and instantiates quantum circuits on near term devices. Cirq incorporates basic QC structures: n n Qubits Gates, circuits Measurement operators Cirq provides simple programming model that abstract Quantum ML applications. Algorithms and solutions evaluated: n n n MNIST Convolutional Neural Networks (QCNNs) gradient of the expectation value of a certain observable Google expanded its Tensorflow ML Fwk for its CIRQ SDK
Google Implements quadrant: QQ Quantum Data n Quantum data are data sources that occur in a natural or artificial quantum system: n n Can be the classical data resulting from quantum mechanical experiments Data generated by a quantum device and then fed into an algorithm as input. Quantum data includes superposition and entanglements, leading to more simple linear joint probability distribution vs exponential amount of classical computational resources and associated distribution model. Hybrid quantum-classical machine learning on “quantum data” could lead to quantum advantage over classical ML Quantum Models n n Google introduced QPU: Quantum Processing Unit The goal is to produce quantum models from quantum datasets
Pace University Student Projects in area of QML n Quantum Computer Vision Tasks – by Raj Ponnusamy n Quantum Computing for Financial Analysis – by Pratik Choudhari n Basic Quantum Computing in Python - by Anthony Escalona n Cross-platform QML experimentation using IBM, Xanadu. AI SDKs and devices n Implementing Shor’s Algorithm on various maps using IBM Q – by Vijay Pawar n Comparison of Quantum Algorithms – by Swapnil Kadekar n Developing an Arabic Encoding Framework for Dialect – by Adnan Alkujuk
- Slides: 12