<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Franck on Medium]]></title>
        <description><![CDATA[Stories by Franck on Medium]]></description>
        <link>https://medium.com/@belemkoabganf?source=rss-1066377d1977------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*FWXp_uKgfghhxNbL.jpg</url>
            <title>Stories by Franck on Medium</title>
            <link>https://medium.com/@belemkoabganf?source=rss-1066377d1977------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 17:04:33 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@belemkoabganf/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Hybrid Quantum-Classical Neural Network and its Application in Handwritten Digit Classification]]></title>
            <link>https://medium.com/mit-6-s089-intro-to-quantum-computing/hybrid-quantum-classical-neural-network-and-its-application-in-handwritten-digit-classification-6c2ac4ec4a31?source=rss-1066377d1977------2</link>
            <guid isPermaLink="false">https://medium.com/p/6c2ac4ec4a31</guid>
            <dc:creator><![CDATA[Franck]]></dc:creator>
            <pubDate>Fri, 29 Jan 2021 17:17:16 GMT</pubDate>
            <atom:updated>2021-01-31T21:02:57.225Z</atom:updated>
            <content:encoded><![CDATA[<p>by Franck N. Belemkoabga and Satvat Jagwani</p><h3>Abstract</h3><p>In this project, we explore and analyze the structure of hybrid quantum-classical neural networks and try training them for solving a toy problem, like the classification of handwritten digits.</p><p>We see that quantum layers are often used to make classically intractable parts of the problem faster and are therefore preceded and followed by classical layers for pre-and post-processing. We also note that although they are able to classify handwritten digits correctly, we need to explore other problems to demonstrate quantum supremacy.</p><h3>Introduction</h3><p>Deep neural networks enabled a wide range of applications in many fields from social to scientific research. For example, speech and facial recognition are examples of deep neural networks present in our phones nowadays. However, deep neural network training is computationally expensive and such limitations can significantly hinder the field. However, new possibilities are becoming available such as the use of quantum computing devices. Quantum Computing devices use the laws of quantum mechanics to overcome the limits of classical computation.</p><h3>The Concept of Neural Network</h3><p>Classical neural networks are composed of units called neurons, which take n-dimensional numerical input, compute a weighted sum of its values, and compute a non-linear activation function of the sum to pass as output. These neurons are arranged in multiple layers, where each layer takes the output of previous units as input and passes its own output to the next unit as input.</p><p>These neural networks are used to compute the output of a machine learning problem given the features of data as input. This is done by a technique called forward propagation, where the result is computed at each layer and passed to the next layer. We also need to train these neural networks, meaning we need to find weights that do the job correctly. This is done by backward propagation, which involves finding the error between desired result and output of the neural network and updating the weights based on how much they contribute to the error.</p><h3>Hybrid Quantum-Classical neural network</h3><p>According to the Qiskit textbook, Hybrid Quantum Classical Neural Networks contain a few classical layers and quantum layers. In general, they have classical layers after the input, followed by a few quantum layers, and then followed by classical layers to give output. For forward propagation, the quantum layers apply unitary operators parameterized by classical layer output on their qubits, and then take a weighted sum of the measured final states to get output. For backward propagation, the gradients for classical layers are calculated the same way as before, and the gradients for quantum layers are calculated using a parameter shift rule.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EfA9YWA2JbgxBa5T" /><figcaption>Figure 1: General procedure of quantum circuit illustrated with how it takes input and how it calculates output. (Fig. 2 in <a href="https://iopscience.iop.org/article/10.1088/2058-9565/ab4eb5">https://iopscience.iop.org/article/10.1088/2058-9565/ab4eb5</a>)</figcaption></figure><p>As noted in <a href="https://arxiv.org/abs/1904.04767">https://arxiv.org/abs/1904.04767</a>, another important feature of many hybrid networks in quanvolutional layers (quantum convolutional layers). These layers apply quantum circuits on spatially local sections of output from previous layers. Since these circuits deal with very small sections of data (as is the nature of convolutions), these have low chances of quantum error. In addition, since we do not know apriori which parameter values are to be used for feature detection, any innate errors with quantum gates do not impact results as these gates are trained for their desired purpose.</p><h3>Application (digit classification)</h3><p>The Qiskit textbook shows how a classical neural network can be partially put in a quantum state to create a hybrid quantum-classical neural network. A combination of Qiskit and Pytorch, an open-source machine learning library, is used to implement a simple example of a hybrid quantum and classical neural network. This model will then be used to create a hybrid quantum-classical neural network that can classify hand-drawn digits.</p><h4>Framework</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*rHY1VmqJtsF8wikB" /><figcaption>Figure 2: Hybrid classical-quantum neural network schema From Qiskit</figcaption></figure><p>This is the framework that will be used to construct the hybrid classical-quantum neural network. As can be seen in the image, there is the classical node supported by PyTorch which then works in conjunction with the Quantum node supported by Qiskit. And finally, the output is handled with PyTorch as well as the updating and optimizing of parameters.</p><p>The code can be accessed by clicking on the link below:</p><p><a href="https://drive.google.com/file/d/16uvm5s3T2plqMO550o1IECJaPNV6hCm_/view?usp=sharing">Hybrid Classical Quantum Neural Network.ipynb</a></p><h3>Conclusion</h3><p>In this project, we explore the functioning of hybrid quantum-classical neural networks and see their application in a classification problem. We see that these hybrid NNs can be used for solving machine learning problems. However, we need to do some more research to demonstrate a Machine Learning application with quantum supremacy.</p><h3>References</h3><p>“Hybrid quantum-classical Neural Networks with PyTorch and Qiskit.” <em>Qiskit</em>, 26 Jan. 2021, qiskit.org/textbook/ch-machine-learning/machine-learning-qiskit-pytorch.html.</p><p>“Parameterized Quantum Circuits As Machine Learning Models.” <em>Error 400 (Bad Request)!!1</em>, 13 Nov. 2019, iopscience.iop.org/article/10.1088/2058–9565/ab4eb5.</p><p>“Quanvolutional Neural Networks: Powering Image Recognition with Quantum Circuits.” <em>ArXiv.org</em>, arxiv.org/abs/1904.04767.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6c2ac4ec4a31" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mit-6-s089-intro-to-quantum-computing/hybrid-quantum-classical-neural-network-and-its-application-in-handwritten-digit-classification-6c2ac4ec4a31">Hybrid Quantum-Classical Neural Network and its Application in Handwritten Digit Classification</a> was originally published in <a href="https://medium.com/mit-6-s089-intro-to-quantum-computing">MIT 6.s089 — Intro to Quantum Computing</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>