Embedded Files
ChemicalQDeviceExisting Quantum Computing Advantages, Neuroimages
Dear Healthcare Colleague, my name is Kevin Kawchak, and I am the Founder CEO of ChemicalQDevice. Today is April 7th, 2023 for "Existing Quantum Computing Advantages for Medical Images/Neuroradiology"
4 Key Articles have highlighted the importance of utilizing emerging quantum technology to train images. These benefits fall under the main category of artificial intelligence to machine learning to deep learning to convolutional neural networks, to Now quantum convolutional neural networks, or quanvolutional neural networks for short.
#1) In a 2018 Vojtech Havlicek, et al. paper titled "Supervised learning with quantum enhanced feature spaces", several quantum computing benefits in regards to classifications were illustrated. The IBM and MIT researchers explained that a Quantum variational classifier could be used to decrease financial expenditures associated with running large feature sets with modern Support Vector Machines.
Quoting the authors' modification of their quantum algorithm implementations "We clearly see an increase in classification success with increasing circuit depth." End Quote In addition, a second approach described combining a quantum kernel estimator to estimate the kernel function when combined with a conventional SVM for classification. [1]
#2) A 2019 article by Maxwell Henderson et al. was titled “Quanvolutional Neural Networks: Powering Image Recognition with Quantum Circuits”, by QxBranch since acquired by Rigetti Computing. The main findings were that Small QNN circuits could be run today on single data points, with no QRAM requirements. In addition, current ML algorithms produce probabilistic results but suffer under computational tasks. Quantum circuits also produce probabilistic results, but have exponential potential and certain algorithms can be run on quantum simulators productively.
The authors highlighted a main difference using quantum computing in that Quanvolutional filters extract features from input images by transforming spatially-local subsections of data. In addition, the number of qubits in the quanvolutional filter does not need to match the dimensionality of the input data, in general. Key questions were proposed which includes: A) Do certain structured quanvolutional filters work better?B) What minimal quanvolutional filter gate depths work better? C) How data-dependent is the ideal “set” of quanvolutional filters? and D) How does encoding and decoding approaches influence results? [2]
#3) A 2019 paper by Iris Cong, et al. titled "Quantum Convolutional Neural Networks" by Harvard University and UC Berkeley presented the quantum computing benefit for images by only using O(log(N)) variational parameters for input sizes of N qubits, compared to unfavorable exponential parameter requirements for classical computing. Quoting the authors “The extreme complexity of many-body states often makes theoretical analysis intractable” End Quote, when using conventional approaches. [3]
#4) And lastly, a 2023 IBM Qiskit model titled “The Quantum Convolution Neural Network” provides a step by step explanation of the quantum computing processes behind training images with convolution and pooling layers, which includes A) Image is encoded into the quantum circuit with feature map B) Kernels are used for determining features and patterns of a particular input C) Pooling layer reduces dimensionality of input, reduces computational cost, and reduces number of learning parameters D) Pooling layer performs operations, then disregards certain qubits E) The Training stage of Convolution and pooling parameters are tuned to reduce the loss function F) Any parametrized circuit should work, but results will vary, and G) The fully connected layer determines classification of input image [4]
At least 7 articles in recent years have been identified that utilize Neuroimages, or other Medical Images combined with principles found in these 4 summarized papers. In other words, hybrid quantum computing will likely play a pivotal role to improve healthcare image analyses. [5-11]
References are available in the comments below. Have a productive rest of your day. Smile.
4/7/23 "Existing Quantum Computing Advantages for Medical Images/Neuroradiology"
References[1] https://arxiv.org/pdf/1804.11326.pdf[2] https://arxiv.org/pdf/1904.04767.pdf[3] https://arxiv.org/pdf/1810.03787.pdf[4] https://qiskit.org/documentation/machine-learning/tutorials/11_quantum_convolutional_neural_networks.html[5] https://aip.scitation.org/doi/10.1063/5.0138021[6] https://www.mdpi.com/2227-7390/11/2/376[7] https://arxiv.org/pdf/2301.12505.pdf[8] https://www.mdpi.com/2079-9292/11/5/721[9] https://arxiv.org/abs/2102.06535[10] https://bmcmedinformdecismak.biomedcentral.com/articles/10.1186/s12911-021-01588-6[11] https://arxiv.org/pdf/2303.10142.pdf
Created by Kevin Kawchak Founder CEO ChemicalQDevice2023 San Diego, CaliforniaHealthcare Innovation
4 Key Articles have highlighted the importance of utilizing emerging quantum technology to train images. These benefits fall under the main category of artificial intelligence to machine learning to deep learning to convolutional neural networks, to Now quantum convolutional neural networks, or quanvolutional neural networks for short.
#1) In a 2018 Vojtech Havlicek, et al. paper titled "Supervised learning with quantum enhanced feature spaces", several quantum computing benefits in regards to classifications were illustrated. The IBM and MIT researchers explained that a Quantum variational classifier could be used to decrease financial expenditures associated with running large feature sets with modern Support Vector Machines.
Quoting the authors' modification of their quantum algorithm implementations "We clearly see an increase in classification success with increasing circuit depth." End Quote In addition, a second approach described combining a quantum kernel estimator to estimate the kernel function when combined with a conventional SVM for classification. [1]
#2) A 2019 article by Maxwell Henderson et al. was titled “Quanvolutional Neural Networks: Powering Image Recognition with Quantum Circuits”, by QxBranch since acquired by Rigetti Computing. The main findings were that Small QNN circuits could be run today on single data points, with no QRAM requirements. In addition, current ML algorithms produce probabilistic results but suffer under computational tasks. Quantum circuits also produce probabilistic results, but have exponential potential and certain algorithms can be run on quantum simulators productively.
The authors highlighted a main difference using quantum computing in that Quanvolutional filters extract features from input images by transforming spatially-local subsections of data. In addition, the number of qubits in the quanvolutional filter does not need to match the dimensionality of the input data, in general. Key questions were proposed which includes: A) Do certain structured quanvolutional filters work better?B) What minimal quanvolutional filter gate depths work better? C) How data-dependent is the ideal “set” of quanvolutional filters? and D) How does encoding and decoding approaches influence results? [2]
#3) A 2019 paper by Iris Cong, et al. titled "Quantum Convolutional Neural Networks" by Harvard University and UC Berkeley presented the quantum computing benefit for images by only using O(log(N)) variational parameters for input sizes of N qubits, compared to unfavorable exponential parameter requirements for classical computing. Quoting the authors “The extreme complexity of many-body states often makes theoretical analysis intractable” End Quote, when using conventional approaches. [3]
#4) And lastly, a 2023 IBM Qiskit model titled “The Quantum Convolution Neural Network” provides a step by step explanation of the quantum computing processes behind training images with convolution and pooling layers, which includes A) Image is encoded into the quantum circuit with feature map B) Kernels are used for determining features and patterns of a particular input C) Pooling layer reduces dimensionality of input, reduces computational cost, and reduces number of learning parameters D) Pooling layer performs operations, then disregards certain qubits E) The Training stage of Convolution and pooling parameters are tuned to reduce the loss function F) Any parametrized circuit should work, but results will vary, and G) The fully connected layer determines classification of input image [4]
At least 7 articles in recent years have been identified that utilize Neuroimages, or other Medical Images combined with principles found in these 4 summarized papers. In other words, hybrid quantum computing will likely play a pivotal role to improve healthcare image analyses. [5-11]
References are available in the comments below. Have a productive rest of your day. Smile.
4/7/23 "Existing Quantum Computing Advantages for Medical Images/Neuroradiology"
References[1] https://arxiv.org/pdf/1804.11326.pdf[2] https://arxiv.org/pdf/1904.04767.pdf[3] https://arxiv.org/pdf/1810.03787.pdf[4] https://qiskit.org/documentation/machine-learning/tutorials/11_quantum_convolutional_neural_networks.html[5] https://aip.scitation.org/doi/10.1063/5.0138021[6] https://www.mdpi.com/2227-7390/11/2/376[7] https://arxiv.org/pdf/2301.12505.pdf[8] https://www.mdpi.com/2079-9292/11/5/721[9] https://arxiv.org/abs/2102.06535[10] https://bmcmedinformdecismak.biomedcentral.com/articles/10.1186/s12911-021-01588-6[11] https://arxiv.org/pdf/2303.10142.pdf
Created by Kevin Kawchak Founder CEO ChemicalQDevice2023 San Diego, CaliforniaHealthcare Innovation
Report abuse