ChemicalQDeviceQML Platform Benefits: Qiskit QML, Pennylane..
Dear Healthcare Colleague, my name is Kevin Kawchak, and I am the Founder CEO of ChemicalQDevice. Today is May 4th, 2023 for "QML Platform Benefits: Qiskit QML, Pennylane, TensorFlow Quantum, and TorchQuantum"
IBM Qiskit Quantum Machiine learning was launched in 2021, and adds implementations based on Qiskit primatives. This includes the NeuralNetwork interface as an abstract class that all Quantum Neural Networks inherit from. Forward and backward passes that take data samples and trainable weights as inputs is also related to NeuralNetwork. NeuralNetwork Regressors and classifiers have an option to plug into a TorchConnector for native use as PyTorch modules in larger PyTorch models.
EstimatorQNN is a network based on the evaluation of quantum observables; and SamplerQNN is a network based on the samples resulting from measuring a quantum circuit. EstimatorQNN and SamplerQNN can also take in any subclass of the classes BaseEstimator and BaseSampler, this according to the Qiskit Quantum Neural Networks page. (1-2)
Xanadu Pennylane was the first QML platform to launch, which was in 2018 to "help control and manipulate parametrized quantum circuits." According to Q-munity, PQC utilization is accomplished with quantum computations referred to as quantum node objects to initialize a quantum circuit. QNodes can interact with classical machine learning libraries such as TensorFlow and Pytorch. In this platform, Qubits are referred to as wires. Quantum circuits used in the Pennylane open source platform are differentiable, which has been a key aspect of classical deep learning. (3-4)
Google TensorFlow Quantum, or TFQ was launched in 2020, and features an advanced qsim simulator that has outperformed Google's Cirq. A technique referred to as Gate fusion performs premultiplication of quantum matrices prior to additional operations. In general, the goals of using TFQ for qml are "to optimize over a parameterized class of computations" to generate certain low energy wavefunctions, #2 learning to extract non-local information, and learning how to generate a quantum distribution from data.
At the top of the TensorFlow Quantum stack begins with either classical data or quantum data. Classical data is processed by TensorFlow, while quantum circuits and quantum operators are processed by TensorFlow Quantum. Next, Keras Models process both classical data which proceeds to TF Layers, and also quantum data which proceeds to TFQ Layers and TFQ Differentiators. TF Ops and TFQ Ops separately instantiatiate dataflow graphs. TFQ qsim simulator processes data by quantum mechanics on a classical computer, and Cirq is generally for creating quantum algorithms. TPUs, GPUs, CPUs, and QPUs can be utilized in these processes, as depicted at the bottom of the stack. (5-6)
MIT TorchQuantum launched early 2022, and has seen a number of developments that focus on fundamentally improving Quantum Machine Learning. The team's goal is to Quote "Leverage algorithm-system co-design methodology, especially with the help of Machine Learning to improve noise-robust, efficiency, and accuracy of quantum circuits on real quantum devices." QuantumNAS, short for Quantum Noise-Adaptive Search, is a comprehensive framework for noise-adaptive co-search of variational circuits and qubit mapping. QuantumNAT, or Quantum Noise-Aware Training, is a Post Quantum Cryptography framework to perform noise-aware optimizations in both training and inference stages for improving robustness. QOC, or Quantum On-Chip Training, is the first experimental demonstration of practical training with parameter shift of this category.
In addition, key manuscripts regarding sQCNN, sQCNN-3D, and PVM have ties to TorchQuantum. The most direct path to improve applications such as medical image analysis will likely be with utilizing these QML platforms on quantum simulators - which have been proven in literature and are noise free, but limited in speed and number of qubits. (7-13) References are available in the comments below. Have a productive rest of your day. Smile.
5/4/23 "QML Platform Benefits: Qiskit QML, Pennylane, TensorFlow Quantum, and TorchQuantum"
References: (1) Qiskit https://qiskit.org/ecosystem/machine-learning/tutorials/01_neural_networks.html(2) Qiskit https://medium.com/qiskit/introducing-qiskit-machine-learning-5f06b6597526(3) Pennylane https://www.qmunity.tech/tutorials/an-introduction-to-pennylane(4) Pennylane https://pennylane.ai/qml/whatisqml(5) TensorFlow Quantum https://www.tensorflow.org/quantum(6) TensorFlow Quantum https://arxiv.org/pdf/2003.02989.pdf(7) TorchQuantum https://qmlsys.mit.edu/ (8) TorchQuantum https://arxiv.org/abs/2107.10845(9) TorchQuantum https://arxiv.org/abs/2110.11331 (10) TorchQuantum https://arxiv.org/abs/2202.13239(11) PVM https://arxiv.org/abs/2210.16731(12) sQCNN https://arxiv.org/abs/2209.12372(13) sQCNN-3D https://arxiv.org/abs/2210.09728
Created by Kevin Kawchak Founder CEO ChemicalQDevice2023 San Diego, CaliforniaHealthcare Innovation
IBM Qiskit Quantum Machiine learning was launched in 2021, and adds implementations based on Qiskit primatives. This includes the NeuralNetwork interface as an abstract class that all Quantum Neural Networks inherit from. Forward and backward passes that take data samples and trainable weights as inputs is also related to NeuralNetwork. NeuralNetwork Regressors and classifiers have an option to plug into a TorchConnector for native use as PyTorch modules in larger PyTorch models.
EstimatorQNN is a network based on the evaluation of quantum observables; and SamplerQNN is a network based on the samples resulting from measuring a quantum circuit. EstimatorQNN and SamplerQNN can also take in any subclass of the classes BaseEstimator and BaseSampler, this according to the Qiskit Quantum Neural Networks page. (1-2)
Xanadu Pennylane was the first QML platform to launch, which was in 2018 to "help control and manipulate parametrized quantum circuits." According to Q-munity, PQC utilization is accomplished with quantum computations referred to as quantum node objects to initialize a quantum circuit. QNodes can interact with classical machine learning libraries such as TensorFlow and Pytorch. In this platform, Qubits are referred to as wires. Quantum circuits used in the Pennylane open source platform are differentiable, which has been a key aspect of classical deep learning. (3-4)
Google TensorFlow Quantum, or TFQ was launched in 2020, and features an advanced qsim simulator that has outperformed Google's Cirq. A technique referred to as Gate fusion performs premultiplication of quantum matrices prior to additional operations. In general, the goals of using TFQ for qml are "to optimize over a parameterized class of computations" to generate certain low energy wavefunctions, #2 learning to extract non-local information, and learning how to generate a quantum distribution from data.
At the top of the TensorFlow Quantum stack begins with either classical data or quantum data. Classical data is processed by TensorFlow, while quantum circuits and quantum operators are processed by TensorFlow Quantum. Next, Keras Models process both classical data which proceeds to TF Layers, and also quantum data which proceeds to TFQ Layers and TFQ Differentiators. TF Ops and TFQ Ops separately instantiatiate dataflow graphs. TFQ qsim simulator processes data by quantum mechanics on a classical computer, and Cirq is generally for creating quantum algorithms. TPUs, GPUs, CPUs, and QPUs can be utilized in these processes, as depicted at the bottom of the stack. (5-6)
MIT TorchQuantum launched early 2022, and has seen a number of developments that focus on fundamentally improving Quantum Machine Learning. The team's goal is to Quote "Leverage algorithm-system co-design methodology, especially with the help of Machine Learning to improve noise-robust, efficiency, and accuracy of quantum circuits on real quantum devices." QuantumNAS, short for Quantum Noise-Adaptive Search, is a comprehensive framework for noise-adaptive co-search of variational circuits and qubit mapping. QuantumNAT, or Quantum Noise-Aware Training, is a Post Quantum Cryptography framework to perform noise-aware optimizations in both training and inference stages for improving robustness. QOC, or Quantum On-Chip Training, is the first experimental demonstration of practical training with parameter shift of this category.
In addition, key manuscripts regarding sQCNN, sQCNN-3D, and PVM have ties to TorchQuantum. The most direct path to improve applications such as medical image analysis will likely be with utilizing these QML platforms on quantum simulators - which have been proven in literature and are noise free, but limited in speed and number of qubits. (7-13) References are available in the comments below. Have a productive rest of your day. Smile.
5/4/23 "QML Platform Benefits: Qiskit QML, Pennylane, TensorFlow Quantum, and TorchQuantum"
References: (1) Qiskit https://qiskit.org/ecosystem/machine-learning/tutorials/01_neural_networks.html(2) Qiskit https://medium.com/qiskit/introducing-qiskit-machine-learning-5f06b6597526(3) Pennylane https://www.qmunity.tech/tutorials/an-introduction-to-pennylane(4) Pennylane https://pennylane.ai/qml/whatisqml(5) TensorFlow Quantum https://www.tensorflow.org/quantum(6) TensorFlow Quantum https://arxiv.org/pdf/2003.02989.pdf(7) TorchQuantum https://qmlsys.mit.edu/ (8) TorchQuantum https://arxiv.org/abs/2107.10845(9) TorchQuantum https://arxiv.org/abs/2110.11331 (10) TorchQuantum https://arxiv.org/abs/2202.13239(11) PVM https://arxiv.org/abs/2210.16731(12) sQCNN https://arxiv.org/abs/2209.12372(13) sQCNN-3D https://arxiv.org/abs/2210.09728
Created by Kevin Kawchak Founder CEO ChemicalQDevice2023 San Diego, CaliforniaHealthcare Innovation