<?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[MyDataModels - Medium]]></title>
        <description><![CDATA[Discover an innovative Augmented Analytics platform accessible to all professional to help them extract value from their data. - Medium]]></description>
        <link>https://medium.com/mydatamodels?source=rss----7901bf652483---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>MyDataModels - Medium</title>
            <link>https://medium.com/mydatamodels?source=rss----7901bf652483---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 23:16:23 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/mydatamodels" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Evolutionary Algorithm for Classification]]></title>
            <link>https://medium.com/mydatamodels/evolutionary-algorithm-for-classification-aafe30fb4222?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/aafe30fb4222</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machinelearningalgorithms]]></category>
            <category><![CDATA[evolutionary-algorithms]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Wed, 09 Jun 2021 12:56:34 GMT</pubDate>
            <atom:updated>2021-06-09T12:56:24.554Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x76XrKTkpxZSH6MnsZYbjw.jpeg" /></figure><p>Mikhail Kamalov is a Ph.D. student with INRIA. We interviewed him for the first time when he published his first scientific paper about <a href="https://www.mydatamodels.com/semi-supervised-learning-for-document-classification/">semi-supervised learning for document classification</a> last November.</p><p>Another one of his articles is accepted for publication this May at a prestigious conference: <a href="https://eusipco2021.org/call-for-papers/">EUSIPCO</a>. We want to learn more about this article and, in particular about his novel use of evolutionary algorithms for classification.</p><p><strong>Hi Mikhail, nice seeing you again. Congratulations on your paper being accepted for EUSIPCO.</strong></p><p>Thanks. We’re happy about it.</p><p><strong>Was it easier this time than for the first article to get a paper accepted?</strong></p><p>I think that it is never easy to get a paper accepted. As scientists, we are humbled by the challenges we face. We had to rework it a couple of times based on the feedback received. But at least, I get better at explaining what I do. It is a great learning experience.</p><h3>Is the topic “Evolutionary algorithm for classification”?</h3><p>It’s more than that. It deals with evolutionary algorithms for classification but not only.</p><h3>What does it mean? Does it deal with data analytics? Predictive analytics? Augmented analytics?</h3><p>Indeed, the whole analytics vocabulary is very trendy, whether data analytics, augmented analytics, or predictive analytics. But as a scientist, these are use cases to me. These are examples of how what I do can be used.</p><p><strong>Do you mean that you do not care about data analytics but only about evolutionary algorithms?</strong></p><p>No, I care about the use cases for my research. But my contribution focuses on the algorithms; the use cases are applications of this work.</p><p><strong>OK, what is your contribution about? Semi-supervised learning algorithms ?</strong></p><p>In part, yes.</p><h3>What is the other part? Genetic algorithms?</h3><p>Yes. It is about both.</p><p><strong>But they have nothing in common. You cannot use both, can you?</strong></p><p>Yes, you can.</p><p><strong>Please explain.</strong></p><p>It is indeed possible to combine algorithms. A technique called “stacking” exists. Model stacking is a form of lego applied to AI models. There are ‘base’ models (i.e., lego blocks), and there is a “superset” model (i.e., the final construct). The base models form the building blocks of the superset structure. At ground 0, there are several or one base model. Their outputs feed the next layer of base models, building a ‘pyramid’ or a pipeline of models. Different AI algorithms compose the base models.</p><p><strong>OK, I got it. You can pipeline together different models, i.e., algorithms. But what’s the point?</strong></p><p>The point is to obtain a performance for the stacked model which is higher than the performance of each model.</p><p><strong>Do you mean that each time you stack two algorithms, the performance of the stacked algorithm is higher than the performance of the individual algorithms?</strong></p><p>It is not systematic, no. Sometimes, the overall performance is higher, and sometimes it is not.</p><p><strong>So what did you stack here?</strong></p><p>We stacked a semi-supervised learning algorithm with a genetic algorithm.</p><p><strong>Why would you do this?</strong></p><p>Because our genetic algorithm, <a href="https://www.mydatamodels.com/tada-predictive-modeling-software/">ZGP</a>, needs as inputs fully labeled data. Each data point entered into ZGP for training has to be ‘tagged’ as belonging to a category, i.e. class. It is thanks to this fully labeled input data that ZGP “learns” about the classification. It is one of the strengths of ZGP. It is good at sorting. It is an evolutionary algorithm for classification (at least in this context).</p><p><strong>And once it has learned, it can classify new data by itself?</strong></p><p>Exactly.</p><p><strong>OK, so why the stacking?</strong></p><p>Because in real situations, we collect data, but it is cumbersome and time-consuming to label each data point manually.</p><p><strong>Oh, you mean that the ‘classification’ is done manually?</strong></p><p>Yes, it is a human being who labels, one by one, each data point in the data set.</p><h3>Is it the same whether you do predictive analytics, data analytics, or augmented analytics?</h3><p><strong>Does it depend on the use case?</strong></p><p>It is time-consuming to label data, whatever the data, whatever the use case.</p><p><strong>OK, so that’s why the semi-supervised learning algorithm?</strong></p><p>Yes, with a semi-supervised learning algorithm such as PRPCA, we can automatically label a whole dataset based on few inputs.</p><p><strong>OK</strong></p><p>Yes, we try to combine the strengths of the semi-supervised classification algorithm PRPCA (the famous algorithm from Google) and the resilience of ZGP.</p><p><strong>So you stack them?</strong></p><p>Yes.</p><p><strong>How does it impact the performances?</strong></p><p>The stacked model shows outstanding performances, much better than PRPCA alone or ZGP alone.</p><p><strong>How do you measure this?</strong></p><p>We took three datasets. We generated the first dataset experimentally. And the two others are publicly available. They are named respectively DC motor (the one we generated), WII and UWave (the two publicly available ones).</p><p><strong>How did you generate the DC motor dataset?</strong></p><p>To work on a real dataset on motor failures, we conducted our experiment to simulate anomalies of DC motors in a production environment. Loading weights generated motor axis imbalance onto a disk plate mounted on top of the motor at varying distances from its axis.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ujZxMAY9IF24aK-JlaOIxA.png" /></figure><p><strong>Interesting, and what did you measure?</strong></p><p>We measured the accelerations, the rotations, and the magnetic fields using an accelerometer, a gyroscope, and a magnetometer.</p><p><strong>Wait for a second. If you want to predict motor failure, then this is a predictive analysis use case!</strong></p><p>True. It is a combination of genetic algorithms, optimization algorithms, and predictive analytics.</p><h3>And you applied the semi-supervised learning algorithm PRPCA stacked with the genetic algorithm ZGP onto the three datasets, i.e., data analytics?</h3><p>Yes, exactly.</p><p><strong>And how were the performances of the stacking?</strong></p><p>Very good. When we apply PRPCA alone onto the DC motor dataset, we obtain accuracies ranging from 66% to 71%.</p><p><strong>OK. That’s pretty good already.</strong></p><p>It’s pretty good, but we can do better.</p><h3>What about the genetic algorithm ZGP alone?</h3><p>ZGP alone has accuracy performances ranging from 26% to 62% on the DC motor dataset.</p><p><strong>OK, and what about when you stack both?</strong></p><p>The stacking of PRPCA (semi-supervised algorithm) and ZGP (genetic algorithm) provides accuracy ranging from 94.2% to 98.8%.</p><p><strong>Excellent!</strong></p><p>That’s probably why the committee in charge of the selection noticed the article.</p><p><strong>And now, this resulting model can be applied to any topic and any use case.</strong></p><p>To a wide variety, yes. It can be used for data analytics, augmented analytics, predictive analytics. It is a semi-supervised evolutionary algorithm for classification.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aafe30fb4222" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/evolutionary-algorithm-for-classification-aafe30fb4222">Evolutionary Algorithm for Classification</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Artificial Intelligence: AI made simple]]></title>
            <link>https://medium.com/mydatamodels/artificial-intelligence-ai-made-simple-c6d91570cda8?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/c6d91570cda8</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Tue, 06 Apr 2021 14:36:12 GMT</pubDate>
            <atom:updated>2021-04-06T14:34:49.620Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8tyyoVpL7folp4tW.jpg" /></figure><p>Let’s start with a few figures:</p><ul><li>AI can increase business <strong>productivity by 40%</strong>.</li><li>Already <strong>77% of the devices</strong> we use feature one form of AI or another.</li><li>AI-powered voice assistants to reach<strong> 8 billion by 2023</strong>.</li></ul><p>Yet, some estimate that artificial intelligence is just another buzzword and that AI, like many trends before, will quickly vanish again. It is time to explain why artificial intelligence AI is much more than just short-term hype. Let’s take a closer glimpse at the topic.</p><p>In 1997, the Deep Blue system, developed by IBM, succeeded in defeating the multiple chess world champion Garri Kimowitsch Kasparov in six games. A computer program that can play chess to this level was unquestionably a breakthrough. The area of artificial intelligence AI at the time hit a new milestone.</p><h3>Definition of Artificial Intelligence AI</h3><p>The term Artificial Intelligence refers to an area of ​​computer science that intends to mimic intelligent human behavior. A computer should therefore be programmed for problem-solving independently without the help of humans.</p><p>It processes perceptions, compares them with previously learned algorithms. It then derives a specific action or recommended action to achieve this goal.</p><h3>How does AI Technology Work?</h3><p>AI systems today essentially consist of neural networks trained with the help of machine learning or deep learning. Such a network is built up of several layers of artificial neurons. Each layer consists of several hundred or a thousand individual neurons.</p><p>Numerous hidden layers are connecting the input and output layers responsible for processing the information.</p><p>When, for example, a neuron from the input layer indicates that a line is curved. A neuron from the following layer can link this information with analogous signals from other neurons from the previous layer. It has then gathered the relevant information to conclude that the object analyzed is a circle.</p><p>This way, more complicated relations are built in each layer. The output layer then understands that a car is clearly depicted on an image or can even identify which car it is.</p><p>Of course, this example is an oversimplified instance of how artificial intelligence operates. The actual processes are much more complex.</p><h3>How do you train artificial intelligence AI for specific tasks?</h3><p>For artificial intelligence to work and differentiate, for example, a car from a bike, the computer ​​systems must first gain the required knowledge.</p><p>In practice, this typically happens through supervised learning. For instance, for computer vision, the AI ​​system must see and learn from countless pictures. These pictures might be of cars or of whatever it is supposed to learn.</p><p>It doesn’t make a difference whether it is pictures, texts, language, or another form of data. It is only essential that the training data record can be treated digitally.</p><p>In the learning process, weighted connections with the neurons of previous layers are also created. They identify relevant and irrelevant neurons of the prior layer. The weights adjust continuously during training until the best achievable outcome is ultimately obtained.</p><h3>Turing Test is not Science Fiction Anymore.</h3><p>The reference test to identify a successful Artificial Intelligence AI algorithm is the Turing test. Here is how it goes.</p><p>On top of the computer, two human test subjects take part in the Turing test. Conversation happens via a keyboard without the test participants having sound or ocular contact.</p><p>One of the test persons talks with both the other test person and the computer. And then this person has to say which of his conversation partners was the machine.</p><p>If this person cannot answer this without hesitation, the machine intelligence passes the Turing test, which is deemed equivalent to humans’.</p><h3>Perception in Artificial Intelligence AI</h3><p>In the term artificial intelligence, there is the notion that a computer should be programmed to resolve problems independently. For this purpose, it has to process perceptions, i.e., input from the real world.</p><h3>Image Processing in Artificial Intelligence AI</h3><p>When recording information using cameras, 3D sensors take pictures of their surroundings. It is also known as computer vision.</p><p>It is currently used in the industry. There, artificial intelligence AI can be used, for example, for monitoring production. It can secure quality assurance, for instance, by checking the surface of a workpiece. It can also carry out a completeness check.</p><p>Image processing is also of use in another category of problem-solving: police and air force security. Indeed, it is possible to train a neural network on millions of pictures of human faces. Once trained, the artificial intelligence AI will be able to recognize individuals. It will be able to identify a male individual from a female one. It will be able to recognize a precise person even though its expression might be different. This exact use case for the usage of AI technology belongs to the category of narrow AI. It means an AI trained on a single, straightforward task.</p><p>Of course, as explained above, the system will be just as performant and efficient as its training data. Here is an example of such a flaw.</p><p>The Chicago police use the Clearview face recognition system to identify potential suspects. It has a database of existing convicted felons. The Chicago police have been sentenced for discrimination against African Americans. Such discrimination was proven to be true. Typically, the Clearview system falsely ‘recognized’ the wrong African American faces. When the issue was investigated further, it appeared as a flaw in the Clearview system. This artificial intelligence AI was trained on a particular task: identifying Caucasian people. When used on African Americans, it reached its limits. It falsely identified felons faces.</p><h3>Voice recognition in Artificial Intelligence AI</h3><p>Voice recognition is another way that computer systems might record information. A person speaks or writes something, and the computer picks up what has been said or written. Again, the quality of the voice recognition depends on the data set used for training.</p><p>This technology is used, for instance, in clever chatbots. They have been used increasingly for communication with customers in recent years and voice assistants for smartphones, such as Siri from Apple.</p><h3>How far is artificial intelligence AI?</h3><p>Many people comprehend the advancement of artificial intelligence AI as a critical menace. It is the fear of independent machines that acquire a life of their own. It could be an intelligent system driving cars. Such an approach might resist the instructions of their developers.</p><p>Nevertheless, the practice seems slightly different. Artificial intelligence still depends on human teachers who feed it data. AI is a front-end that transforms the vast amount of data that big data brings into usable data.</p><p>But it can only draw the correct conclusions from the data if it has beforehand gained the appropriate experience.</p><p>AI is currently working fine when it comes to solving specific tasks in a particular area. However, it cannot now resolve elaborate problems from different departments simultaneously.</p><p>The AI research is working in this direction. Yet, it will still take a while until this time has come — the time when AI may even be equal to human intelligence.</p><p>Currently, amongst other things, the present hardware in the data centers is a barrier for the further growth of artificial intelligence AI. It was the leading cause for the AI winter experienced between 1990 and 2010. Due to the ever-increasing data records of AI applications, even large database systems will ultimately reach their limits.</p><h3>Providing more Resources to AI Cloud Computing</h3><p>Cloud computing is a ‘platform as a service. This technology lets users collate and share a pool of computing resources. Rather than investing in vast coalitions of hardware, office stations and ensuring connectivity with the data centers, a small business can rent rack space, maybe even lease unused CPU cycles. The most popular cloud computing platforms are Amazon, Salesforce, Google, IBM, and Zoho.</p><h3>Conclusion on Artificial Intelligence AI</h3><p>Therefore, one of the upcoming challenges will be implementing flexible and scalable software platforms. It must sustain the interaction between artificial intelligence and big data and further develop this promising technology.</p><p>A distinction is made between strong AI (multi-department problem solving) and weak AI (mom-department problem solving) in artificial intelligence AI. Weak AI is already beating humans in some tasks, or at least is on an equal footing.</p><p>To speak of strong AI, a corresponding system in all areas would have to reach the level of the human brain.</p><h3>References</h3><p><a href="https://www.forbes.com/sites/bernardmarr/2017/04/25/the-complete-beginners-guide-to-artificial-intelligence/?sh=1284c7494a83">https://www.forbes.com/sites/bernardmarr/2017/04/25/the-complete-beginners-guide-to-artificial-intelligence/?sh=1284c7494a83</a></p><p><a href="https://dficlub.org/what-is-artificial-intelligence-guide-for-dummies/">https://dficlub.org/what-is-artificial-intelligence-guide-for-dummies/</a></p><p><a href="https://techjury.net/blog/ai-statistics/">https://techjury.net/blog/ai-statistics/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c6d91570cda8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/artificial-intelligence-ai-made-simple-c6d91570cda8">Artificial Intelligence: AI made simple</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Small Data + Unknown Data Distribution = Machine Learning?]]></title>
            <link>https://medium.com/mydatamodels/small-data-unknown-data-distribution-machine-learning-9f42bf30a434?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/9f42bf30a434</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[small-data]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Thu, 11 Mar 2021 10:13:56 GMT</pubDate>
            <atom:updated>2021-03-11T10:13:43.935Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*i6h20_rpGIL1XEba.jpg" /></figure><ul><li>A cancer researcher has data about the genotype of <a href="https://www.mydatamodels.com/solutions/tada-helps-diagnose-uterus-cancer-based-on-genes/">90 patients</a> and wants to understand who might develop cancer.</li><li>A marketing manager has gathered Data on about 20 000 customers, among which 100 have <a href="https://www.mydatamodels.com/topics/churn-prediction/">churned</a>.</li></ul><p>These are examples of <strong>small datasets with no known data distribution available to professionals</strong>. Do the use of small data sets without a known data distribution mean no possibility to use Machine Learning?</p><p>At MyDataModels, we have designed a <strong>Machine Learning Platform, </strong><a href="https://www.mydatamodels.com/tada-predictive-modeling-software/"><strong>TADA</strong></a>. It can be used by anybody and performs well on small data sets without a known data distribution. In this article, we’re going to share with you how we’re doing it.</p><p>We have no prior knowledge of the type of data used. Still, we believe that our users all deserve to have access to a powerful data analysis tool. That’s why the first pillar of our approach is <strong>to turn data into formulas, i.e., use symbolic regression.</strong></p><h3>Symbolic Regression</h3><p>We try to explore the scope of all possible <strong>mathematical</strong> <strong>formulas</strong>. We look for the ones that best predict the output variable using as input the input variables. We begin from a set of base functions like addition, multiplication, trigonometric functions, and square root. Even though it is challenging to create symbolic models, they have some v<strong>ery beneficial features.</strong></p><p>For starters, a <strong>symbolic</strong> <strong>model</strong> <strong>is explicit</strong>, making it understandable and providing insight into the data. It is also simple, given that th<strong>e optimization process will actively try to keep the formulas as short as possible</strong>. From a technical point of view, a symbolic model is very portable. Anyone can quickly implement it in any programming language without the need for complex data structures. <strong>TADA provides the code for the formulas generated in C++, Java and Python.</strong></p><h3>Global Optimization Under Constraints</h3><p>Still, assuming that we do not know the data distribution at hand, whether a random variable or an exponential distribution. <strong>We use the concepts of global optimization under constraints, our second pillar.</strong> Why?</p><p>Because as a constraint, we choose to limit the complexity of the formulas generated. The goal of global optimization is to attain the globally best solution of models in the presence of multiple local optima. A constraint is a hard limit placed on a variable, which prevents us from going forever in specific directions.</p><h3>Non-Parametric Approaches</h3><p>The third pillar consists in the <strong>explicit use of non-parametric approaches</strong>. Parametric approaches assume a known data distribution, for instance, normal distribution of values or a “bell-shaped curve.” For example, weight is roughly a normal distribution. If you were to graph weight from a population, one would observe a typical bell-shaped curve.</p><p>We use non-parametric approaches in cases where parametric methods, i.e., Poisson distribution, are not appropriate. Non-parametric techniques can often be as potent as parametric ones.</p><h3>Evolutionary Algorithms</h3><p>When doing data science over unknown data distribution, we can use <a href="https://www.mydatamodels.com/a-short-history-of-evolutionary-algorithms">evolutionary algorithms</a>.</p><p>In a nutshell, <strong>how come TADA is so performant on small datasets?</strong></p><p>It uses symbolic regression, combined with global optimization under constraints and a non-parametric approach to its data distribution. All this, under the umbrella of its evolutionary algorithms.</p><h4>References</h4><p><a href="https://www.tno.nl/en/focus-areas/artificial-intelligence/challenges/learning-from-small-datasets/">https://www.tno.nl/en/focus-areas/artificial-intelligence/challenges/learning-from-small-datasets/</a></p><p><a href="https://towardsdatascience.com/why-small-data-is-the-future-of-ai-cb7d705b7f0a">https://towardsdatascience.com/why-small-data-is-the-future-of-ai-cb7d705b7f0a</a></p><p><a href="https://hbr.org/2019/01/the-future-of-ai-will-be-about-less-data-not-more">https://hbr.org/2019/01/the-future-of-ai-will-be-about-less-data-not-more</a></p><p><a href="https://towardsdatascience.com/symbolic-regression-the-forgotten-machine-learning-method-ac50365a7d95">https://towardsdatascience.com/symbolic-regression-the-forgotten-machine-learning-method-ac50365a7d95</a></p><p><a href="https://towardsdatascience.com/symbolic-regression-the-forgotten-machine-learning-method-ac50365a7d95">https://towardsdatascience.com/symbolic-regression-the-forgotten-machine-learning-method-ac50365a7d95</a></p><p><a href="https://mathworld.wolfram.com/GlobalOptimization.html">https://mathworld.wolfram.com/GlobalOptimization.html</a></p><p><a href="https://web.stanford.edu/group/sisl/k12/optimization/MO-unit3-pdfs/3.1introandgraphical.pdf">https://web.stanford.edu/group/sisl/k12/optimization/MO-unit3-pdfs/3.1introandgraphical.pdf</a></p><p><a href="https://ellistat.com/fr/guide-dutilisateur/statistique-inferentielle/tests-parametriques-vs-non-parametriques">https://ellistat.com/fr/guide-dutilisateur/statistique-inferentielle/tests-parametriques-vs-non-parametriques</a></p><p><a href="https://www.sciencedirect.com/topics/medicine-and-dentistry/parametric-test">https://www.sciencedirect.com/topics/medicine-and-dentistry/parametric-test</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f42bf30a434" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/small-data-unknown-data-distribution-machine-learning-9f42bf30a434">Small Data + Unknown Data Distribution = Machine Learning?</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Short History Of Evolutionary Algorithms]]></title>
            <link>https://medium.com/mydatamodels/a-short-history-of-evolutionary-algorithms-1df1926a9a58?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/1df1926a9a58</guid>
            <category><![CDATA[algorithms]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[evolutionary-algorithms]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Mon, 22 Feb 2021 16:08:27 GMT</pubDate>
            <atom:updated>2021-02-22T16:08:27.754Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*A5k-irwFCMjsLdKJ.jpg" /></figure><p>Human beings once had a queue like dogs and cats. This trait has disappeared over time. Nowadays, it survives as a slightly more prominent bone at the bottom of the spine for some of us.</p><p>17% of the world population has blue eyes. In occidental countries, between 10 and 15% of people are dominantly left-handed.</p><p>All of these are genetic characteristics that result from years of biological evolutions.</p><h3>Natural Selection</h3><p>According to Darwin, all species arise and develop through a selective process. It is the natural selection of small, inherited variations that increase the individual’s ability to compete, survive, and reproduce. Those who survive do so through a process called ‘Natural selection.’</p><p>And this same process applies to evolutionary algorithms.</p><p>Natural selection starts with the selection through the survival of the fittest elements belonging to a population. They generate descendants that inherit the features of the parents. These features, in turn, will be added to the next generation.</p><p>A <strong>genetic algorithm</strong> (GA) is a search heuristic inspired by Charles Darwin’s natural evolution theory. This type of evolutionary algorithm reflects the process of natural selection.</p><p>In this context, the fittest individuals (algorithms) are selected for reproduction. The goal is to produce fitter offsprings for the next generation. Genetic algorithms are a subset of evolutionary algorithms.</p><p>Natural selection for humans refers to some individuals’ survival and reproduction due to differences in phenotype, i.e., genes and their expressions. It is a crucial mechanism of biological evolution. Some changes in the traits of a population inherited from several generations carry over to the next.</p><p>Indeed, variations exist within all populations of living beings. It occurs because random mutations arise in an individual organism’s genome, and their descendants can receive such mutations.</p><p>During the lives of the people, their genomes interact with their surroundings to induce variations. And individuals with specific variants of a trait tend to survive and reproduce better than other people with different, less successful variants. As a consequence, the population evolves.</p><p>The premise of an <a href="https://www.mydatamodels.com/interview-with-a-senior-researcher-in-machine-learning-dr-marc-schoenauer/">evolutionary algorithm</a> (<strong>EA</strong>) is quite accessible, given that you are familiar with the process of natural selection. An evolutionary algorithm contains four overall steps: <strong>initialization, selection, genetic operators, and termination</strong>.</p><h3>Population and Gene Selection</h3><p>Natural selection favors some phenotypes. Over long periods of evolution, this process might result in populations specializing in specific biological niches. As a consequence, it may ultimately end in the appearance of new species. In other words, natural selection is an essential process in the biological evolution of a population.</p><p>If parents are fit, their offsprings will be fitter and have a better chance at surviving. This process keeps on iterating. And in the end, we will obtain a generation with the fittest individuals.</p><p>We wish to find the best combination of elements composing the algorithm that maximizes the resulting algorithm’s performance. We will accept an optimal end solution in two instances. Either once we have ran the algorithm for some maximum number of iterations or reached a given performance level. This approach is not the only way to optimize a genetic and evolutionary algorithm, including many typical applications.</p><p>How do we evaluate the performance of an evolutionary algorithm? We use a function to measure the algorithm’s performance objectively. It is called a fitness function.</p><p>It is a particular objective function. It summarises how close a given solution is to achieving the set aims. Fitness functions serve in genetic programming and genetic algorithms to evaluate an offspring’s distance to an optimal solution.</p><p>Now let’s take a look at genetic algorithms. A ‘Genetic Algorithm’ (GA) is a search-based optimization technique. It can find optimal solutions to ‘difficult’ problems, otherwise long to solve.</p><p>A “pool” or a “population” of possible solutions to a given situation is generated randomly. A fitness function is computed for each randomly generated function.</p><p>Some of the solutions generated during this first generation will become parents for the next generation. Therefore we perform a parent selection mechanism among the population.</p><p>It allows breeding, also known as recombination (“crossover”), to produce new children (“the offspring”). In this candidate parent population, each individual (or candidate solution) is assigned a fitness function. The fitter individuals are given a higher chance to mate and generate more “fitter” individuals…(“recombination or crossover”).</p><p>A mutation is a random variation imposed on some individuals. It contributes to additional ‘variability.’ As a result of both breeding and mutation, a new generation of solutions is generated. The same approach is used: selection, breeding/crossover, mutation.</p><p>At each generation, a fitness function is computed for each member of the generation. The processing stops in two cases. When we have generated a given number of generations or reach a predefined performance threshold (fitness).</p><h3>An Example of Evolutionary Algorithms</h3><p>Let’s take an analogy. We want to create a 3D model of the earth using genetic and evolutionary algorithms. Initially, we generate random 3D shapes, say three cubes: a small one, a medium one and a big one and a sphere: a small one, a medium one and a big one, a star: a small one, a medium one, and a big one.</p><p>The fitness function consists in making the difference between the actual 3D model of the earth and those nine initial models. Let’s assume further that the three spheres are retained to breed. They also generate ten spheres of variable sizes.</p><p>A mutation is imposed on the spheres, which add the color green to some of them. In this second generation, the fitness function selects the biggest spheres with the green color mutation (because it approached the earth’s blue).</p><p>Crossover (or breeding) happens between the chosen parents. In the next generation, only giant spheres are generated. A mutation is introduced, which transforms the green spheres into blue ones.</p><p>At this stage, one of the offsprings may be a giant enough sphere of blue color close enough to a 3D representation of the earth. The algorithm stops. We have a model for generating a 3D version of the planet. And we can use this model to ‘predict’ the future state of the planet.</p><p>Evolutionary algorithms are ideally suited for optimization, classification, prediction, and evaluation problems. As a subset of Machine Learning, itself a subset of Artificial Intelligence, this optimization method is efficient in solving complex problems. It is quite an old discipline in computer science.</p><p>Genetic Algorithms (GA) and Evolutionary Strategies were defined in the 70s, while Genetic Programming appeared in the early 90s. TADA, the Machine Learning platform from MyDataModels, provides all the sophistication and the complexity of evolutionary algorithms packaged in an easy-to-use tool.</p><h4>References :</h4><p><a href="https://www.researchgate.net/publication/216300863_A_history_of_evolutionary_computation">https://www.researchgate.net/publication/216300863_A_history_of_evolutionary_computation</a></p><p><a href="https://towardsdatascience.com/introduction-to-evolutionary-algorithms-a8594b484ac">https://towardsdatascience.com/introduction-to-evolutionary-algorithms-a8594b484ac</a></p><p><a href="https://towardsdatascience.com/how-to-define-a-fitness-function-in-a-genetic-algorithm-be572b9ea3b4">https://towardsdatascience.com/how-to-define-a-fitness-function-in-a-genetic-algorithm-be572b9ea3b4</a></p><p><a href="https://towardsdatascience.com/introduction-to-genetic-algorithms-including-example-code-e396e98d8bf3">https://towardsdatascience.com/introduction-to-genetic-algorithms-including-example-code-e396e98d8bf3</a></p><p><a href="https://en.wikipedia.org/wiki/Darwinism#:~:text=Darwinism%20is%20a%20theory%20of,compete%2C%20survive%2C%20and%20reproduce.">https://en.wikipedia.org/wiki/Darwinism#:~:text=Darwinism%20is%20a%20theory%20of,compete%2C%20survive%2C%20and%20reproduce.</a></p><p><a href="https://www.healthline.com/health/red-hair-blue-eyes">https://www.healthline.com/health/red-hair-blue-eyes</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1df1926a9a58" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/a-short-history-of-evolutionary-algorithms-1df1926a9a58">A Short History Of Evolutionary Algorithms</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI At My Service]]></title>
            <link>https://medium.com/mydatamodels/ai-at-my-service-ddce839f4d4b?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/ddce839f4d4b</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Mon, 22 Feb 2021 16:05:35 GMT</pubDate>
            <atom:updated>2021-02-22T16:05:23.553Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3v9i2gCgACn4eoBk.jpg" /></figure><ul><li>A marketing manager for an e-business wants to reduce his churn.</li><li>A meteorologist wants to know when the next pollution peak will happen.</li><li>A cancer researcher wants to find out which genes out of a pool of 53 000 are critical in discriminating the risk of uterus cancer in a woman.</li></ul><p>What do they have in common? They do different jobs in different industries.</p><p>Yet, <strong>they all gather data from their daily activities</strong>. And they all store them in spreadsheets. Whether it is excel based or tableau based.</p><p>And they need to <strong>extract meaningful insights from their data</strong>. The marketing manager needs to understand why some <a href="https://www.mydatamodels.com/solutions/improving-customer-retention-with-churn-prediction/">customers churn</a> and others don’t. The meteorologist wants to know when a <a href="https://www.mydatamodels.com/solutions/predict-and-prevent-pollution-peaks-with-ai/">pollution peak</a> is likely to issue a warning to the population. The cancer researcher wants to find <a href="https://www.mydatamodels.com/solutions/tada-helps-diagnose-uterus-cancer-based-on-genes/">which genes are relevant to her research among all 53 000</a>.</p><p>These are typical uses of artificial intelligence. But none of them is a data scientist nor an AI researcher.</p><p>That’s why at MyDataModels, we have designed <a href="https://www.mydatamodels.com/tada-predictive-modeling-software/">TADA</a>. To make the uses of artificial intelligence available to anyone. You can upload a spreadsheet into the tool in a few clicks, ask it to predict a variable based on this spreadsheet, and get a meaningful result. <strong>We have designed TADA with simplicity of use in mind.</strong></p><p>However, it is both <strong>simple to use and extremely powerful</strong>. The algorithms hidden behind the easy user interface are the results of ten years of research. They are patented. Whatever the AI technologies used behind the scene, whether neural networks or genetic algorithms,<strong> TADA appears simple for its users.</strong></p><p>Because we, at MyDataModels, have worked hard to hide its complexity and <strong>provide you with best in class Machine Learning algorithms</strong>. It allows you to expand your use of artificial intelligence to any data stored in a spreadsheet. TADA is the opposite of a complex wild artificial intelligence tool. <strong>It is a tamed AI at your service.</strong></p><h3>What can you do with it?</h3><p>Here are various uses of artificial intelligence:</p><p><a href="https://www.mydatamodels.com/goals/detection/"><strong>Detection</strong></a><strong>:</strong> For instance, a doctor has gathered numerous pieces of information about a patient’s physical characteristics and symptoms. TADA can identify for each patient, whether they are sick with arthritis, AIDS, or malaria.</p><p><a href="https://www.mydatamodels.com/goals/optimization/"><strong>Optimization</strong></a>. An architect wants to optimize the heating load for a new house to build. TADA can help him choose the right amount of wall and windows surface.</p><p><a href="https://www.mydatamodels.com/goals/prediction/"><strong>Prediction</strong></a>. Our meteorologist wants to predict whether a pollution peak will occur tomorrow. TADA can do it for him.</p><p><a href="https://www.mydatamodels.com/goals/valuation/"><strong>Valuation</strong></a>. A car dealer wants to sell a used car without wasting any time. Hence he needs to find the right market price. He enters his historical data into TADA, and TADA provides him with an objective valuation for the car.</p><p>And in all these cases, TADA explains its results to its users. It explains <strong>which criteria it uses to build a prediction</strong> and how critical each criteria is in a simple bar chart.</p><p>Nothing wild or complicated about TADA. It is a <strong>virtual assistant in disguise for anybody using spreadsheets</strong>. It does not require large amounts of data to work well. A robust machine learning algorithm tamed and available behind an <strong>easy-to-use user interface </strong>for anyone to use. TADA: tamed AI at my service.</p><p><a href="https://www.mydatamodels.com/get-tada/">Get started with TADA!</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ddce839f4d4b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/ai-at-my-service-ddce839f4d4b">AI At My Service</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Love at First Sight is Outdated, Try Machine Learning Engineered Love]]></title>
            <link>https://medium.com/mydatamodels/love-at-first-sight-is-outdated-try-machine-learning-engineered-love-28c34350634a?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/28c34350634a</guid>
            <category><![CDATA[engineered-love]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Fri, 12 Feb 2021 15:42:19 GMT</pubDate>
            <atom:updated>2021-02-12T15:41:52.590Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qg4MWta7aJrSLfpxCi5DnA.jpeg" /></figure><p>After we gave up hope on a family member aged forty who had, to our knowledge (which might not be exhaustive), been single for most of his life, he introduced us to his new girlfriend. A match made in heaven. They had met online. One year later, they were pregnant with a baby girl. Add one more year, and they were happily married. Another year and they were moving into their apartment in the center of Paris.</p><p>I believed this was a rare case of good matchmaking through an online website. And I discovered (fifteen years into their wedding and three baby boys later) that one of my brothers had met his wife online. As the examples continued to pop up around us, I wondered what the global figures were for online dating.</p><p>In 2015, 50% of the U.S. population consisted of single adults. Today, 40% of Americans use online dating. 20% of current, committed relationships began online. It seems that people turn to online dating for a variety of reasons. 48% do it for fun. Some look for romance, and one-in-ten are merely looking for sex (13%). People share information with others too quickly when they are dating online, with a quarter (25%) recognizing they share their full name publicly on their dating profile. One-in-ten have shared their home address, and the same number have shared naked photos of themselves this way, exposing them to risk.</p><p>Numerous surveys estimate that approximately 70% of the couples will meet online in 2040. And no doubt, technology will play a significant role in the matchmaking process. Some dating apps such as AIMM use AI to investigate the character of participants and match them accordingly. It includes their sexual preferences.</p><p>There is no doubt that TADA, from MyDataModels, can be used as the Machine Learning engine for a dating app. It can be trained with historical data from previous successful and unsuccessful matches and predict the best matches.</p><p>Is the use of Machine Learning for matchmaking a good or a bad thing? We cannot give a definite answer. Indeed, some of the old school romantics out there will yearn for the long gone times when you could simply approach a person on the street and invite them out for a coffee. And we all hope that once the COVID pandemic is under control, this kind of romantic encounter can happen again. However, COVID times mean fewer opportunities for random encounters and an accelerated use of online dating.</p><p>Will Machine Learning engineered love resemble old time love? What does old time Love look like?</p><p>Back in California, one of my single French friends told me he was looking for a life partner. An engineer by training, he had a set of specifications for her in mind. She had to be a female with blue eyes, in the same age range as him, and share the same religious faith because “It is such a huge challenge to make a marriage successful.”</p><p>Fast forward twenty years later. This friend found a companion with blue eyes in his age range. They share the same faith and have two children. Right on target! A perfect match on paper. Yet, each time we meet, I walk out with the feeling he lives in an abusive relationship. And I do not dare to share this feeling with him. Matching a set of specifications might not be all there is. And such a potent match does not necessarily mean love nor mutual respect.</p><p>I married a man I once hated and who once hated me. We’ve been married since 1996 (please do not ask me to count the years) to our awe. We’ve made an unspoken compromise not to count nor to celebrate our wedding anniversaries since 2016.</p><p>One of our Caucasian friends had an African girlfriend who became his wife. She was a beautiful woman. Each time someone saw a picture of her, he would comment, “Don’t ask, I don’t know how I seduced her.” with the malicious half-smile of inconspicuous victory on his face. At their wedding, his father explained to whoever was willing to listen that: “There is no such thing as racism with beautiful women.” This saying cheers me up with silent laughter each time I think of it, to this day. It sounds like a perfect way to greet and welcome a daughter-in-law in his family, granting her unconditional love, regardless of skin color or place of birth. She’s beautiful. His son loves her. What else?</p><p>So what works for finding a soul mate? A rational strategy? Full randomness? Is it possible to engineer love?</p><p>Would my friend be in a possibly abusive relationship had he been matched by a Machine Learning algorithm to his wife-to-be?</p><p>Might a Machine Learning algorithm have matched two people who disliked each other at first sight?</p><p>Might a Machine Learning algorithm have introduced a young woman from Africa to a young man from Grenoble in France (a place where it snows half of the year)?</p><p>Maybe. I believe it’s possible. As long as we do not train the Machine Learning algorithms with our own biases, they might well be capable of proposing unlikely yet potentially successful matches.</p><p>And on this Valentine day, let me share one of my favorite quotes from Kavita Daswani in her book I love on romance and wedding, “For Matrimonial Purposes”: <em>“Your life partner is already born. You two just need to find each other.”</em></p><h4><strong>References:</strong></h4><p><a href="https://livre.fnac.com/a1890278/Kavita-Daswani-Mariage-a-l-indienne">https://livre.fnac.com/a1890278/Kavita-Daswani-Mariage-a-l-indienne</a><br><a href="https://www.goodreads.com/book/show/127998.For_Matrimonial_Purposes">https://www.goodreads.com/book/show/127998.For_Matrimonial_Purposes</a><br><a href="https://becominghuman.ai/ai-for-dating-apps-how-machines-help-people-find-love-526ee1088923">https://becominghuman.ai/ai-for-dating-apps-how-machines-help-people-find-love-526ee1088923<br>https://becominghuman.ai/ai-for-dating-apps-how-machines-help-people-find-love-526ee1088923</a><br><a href="http://www.stevierayentertainment.com/fun-wedding-facts/#:~:text=MORE%20WEDDING%20STATISTICS,150%2C000%20per%20year%20in%20Canada).&amp;text=The%20average%20amount%20spent%20on%20a%20traditional%20American%20wedding%20is%20%2422%2C000">http://www.stevierayentertainment.com/fun-wedding-facts/#:~:text=MORE%20WEDDING%20STATISTICS,150%2C000%20per%20year%20in%20Canada).&amp;text=The%20average%20amount%20spent%20on%20a%20traditional%20American%20wedding%20is%20%2422%2C000</a>.<br><a href="https://www.kaspersky.com/blog/online-dating-report/">https://www.kaspersky.com/blog/online-dating-report/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=28c34350634a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/love-at-first-sight-is-outdated-try-machine-learning-engineered-love-28c34350634a">Love at First Sight is Outdated, Try Machine Learning Engineered Love</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beards and Startups: a woman’s perspective]]></title>
            <link>https://medium.com/mydatamodels/beards-and-startups-a-womans-perspective-b0ebcbe9b1fe?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/b0ebcbe9b1fe</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[startup-life]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Thu, 31 Dec 2020 15:42:06 GMT</pubDate>
            <atom:updated>2020-12-31T15:40:26.407Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Toif5S5paCISJKaJe9ksFg.jpeg" /></figure><p>I have recently noticed a strong correlation between men growing beards and men working in startups. I cannot think of a single startup where no bearded person is working. And it makes me wonder: how come? Why would someone who works in a startup grow a beard?</p><p>Therefore I investigated. Is there any correlation (proven or imaginary) between beards and success? After all, it makes sense. What is the goal of a startup employee? Why would someone take on a risky venture, with a lower salary than in a big corporation, in exchange for a super-power title (CxO or VP of something) and a stack of stock options that might (or might not) be worth something one day?</p><p>One possible reason might be ‘because they reach out to be successful.’ I mean, not successful as in ‘If you perform, you will regularly get promoted.’ Successful as in ‘Wow, we are a handful, and we managed to make a product which sells so well that we can hardly follow up.’ Successful like your world is spinning out of control.</p><p>Could it be the hunt for success which drives startup workers to grow beards? Hummmm… .let’s see. If we take one step back and look at the history lessons we attended throughout our education. Does history teach us that bearded men are successful? Maybe so. Zeus, the father of gods in ancient Greece, was the super-mighty. He was quite successful. Lincoln was a courageous US president, and he was very successful too, until he was shot to death. Not an excellent way to finish one’s life, though.</p><p>What about more recent role models? Bill Gates is not bearded. But Steve Jobs and Steve Wozniack both were. That’s it! Eureka! That’s why people who work in a startup grow a beard. Because one of the first super successful startups, born and raised in the heart of Silicon Valley, was founded by two bearded entrepreneurs. Apple is a role model company for many of us in the startup world. So no wonder that its founders also act as role models for us. Apple was the first company ever to ‘share’ stocks with all its employees before going public, an idea from Steve Wozniack. A while back, I had the opportunity to chat with an assistant manager who had worked there in the company’s early stages. She recalled that it was usual to cross the big bosses walking shoeless in the corridors while discussing together.</p><p>So, as a female worker in a startup, how do I feel about the bearded men? I think I don’t know their motivations. And I think it does not matter so much. Contributing to the adventure of a startup is not about making a lot of money quickly. It is not about becoming famous. It is about pursuing a dream and reaching out to become part of a myth. The kind of adventure that makes you feel good about yourself when you look back at your life. It is about having an impact, making a difference. I am confident that my male colleagues, the bearded ones and the others altogether, want to be part of a story bigger than themselves, just as I do. And if growing a beard makes them feel like they are a little closer to being part of the myth, a little closer to the Steves from Apple, I respect that.</p><p>So, how are we doing at MyDataModels? <a href="http://www.mydatamodels.com/company/team">We count 12 people out of 30 who wear a beard</a>. Denis, Marcello, Raymond, Martin, Carlo, Claude, Vincent, David, Alexandre, Jonathan, Mikhail, and Alexis, we welcome your beards, dreams, and energy.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0ebcbe9b1fe" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/beards-and-startups-a-womans-perspective-b0ebcbe9b1fe">Beards and Startups: a woman’s perspective</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Multiclass Classification: Sorting Algorithms]]></title>
            <link>https://medium.com/mydatamodels/multiclass-classification-sorting-algorithms-2fa8f76e37e7?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/2fa8f76e37e7</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Fri, 06 Nov 2020 08:55:54 GMT</pubDate>
            <atom:updated>2020-11-06T08:55:16.861Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wKBnjJ-wCygBqDKO.jpg" /></figure><p>Sorting algorithms are to Machine Learning what the sorting hat is to students in the Harry Potter series: a way to assign each individual to a class (or a house). In Harry Potter, there were four different houses: Gryffindor, Slytherin, Ravenclaw, and Hufflepuff. The magical sorting hat assigns every first-year student into one of the four houses. The sorting hat ‘classifies’ the students into four different categories.</p><p>We have previously discussed <a href="https://www.mydatamodels.com/how-good-is-your-machine-learning-algorithm">binary classification</a>. When the number of classes exceeds two, the terminology used is ‘multiclass’ classification.</p><p>We explained the various metrics used in binary classification, namely: the confusion matrix, Accuracy, Recall, Precision, MCC, F1 score. All these metrics apply to multiclass classification. Let’s see how.</p><h3>Usual Suspects in Sorting Algorithms Metrics: Accuracy and Co.</h3><p>Let’s consider a group of freshmen and assume that we have built a model to predict their house assignment. We want to evaluate the Accuracy of this prediction versus the actual house assignment. Let’s assume further that there are three houses (i.e., three classes): house G, house R, house H. Suppose we get the following confusion matrix:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gBkkg-S-41_5W8EQ.png" /></figure><p>The diagonal represents the number of individuals for which the prediction and the actual house choice match.</p><p>The Accuracy is the ratio of correct classifications over the total. Here:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nrzYcFEjkTkKDlq_.png" /></figure><p>The accuracy, in this case, is 74%.</p><p>The Recall is the number of True Positives (TP) over the total number predicted for one class, i.e., True Positives and False Negatives (FN). But in a multiclass configuration, the Macro-Recall becomes the average of the Recall for each category. The Recall for each of one the different houses are:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*TjCQIVn2HkJ0F3NR.png" /></figure><p>The Macro-Recall is 67%.</p><p>The Precision is calculated in the same way: TP over TP plus False Positives (FP). First, it is computed per class, and then the Macro-Precision is averaged over the three categories.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*MRWi5qQ3c5pefi1q.png" /></figure><p>Metrics are explained here with three classes, and can be generalized to n classes whatever the number of elements to classify.</p><p>The F1 score is known as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*foevwzRjQdXNjyee.png" /></figure><p>The same approach is used for the F1-score. It is calculated for each house (i.e., class) and averaged over the three classes to get the Macro-F1.</p><p>In our case of sorting students into houses, we would get the following values:</p><p>HouseRecallPrecisionF1-score H0.460.70.55G0.860.780.79R0.80.690.74</p><p>As a consequence:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NEwZ19PzyAiPvRI0.png" /></figure><p>So far, the metrics used to evaluate sorting algorithms are familiar. However, there is another metric used specifically for multiclass classification, the Kappa score.</p><h3>The Kappa Score (aka Cohen’s Kappa Coefficient)</h3><p>The Kappa score was initially established in the field of psychology. It is used for measuring the Agreement between two experts (i.e., psychologists) when rating subjects (patients). The Machine Learning community adopted it to estimate classification performance. It has since been used often in computer science.</p><p>Here is an example of the Kappa score as specified at first. Let’s suppose we attend a culinary contest. Two expert chefs, Chef A and Chef B, are judging the dishes cooked by the contestants. The chefs can rank each recipe as ‘Exquisite,’ ‘No,’ ‘Maybe.’ The two chefs might agree on some dishes and disagree on some others.</p><p>The more compatible their ratings, the higher is their ‘Agreement.’ A high level of Agreement confirms that the grades are consistent and make sense, while a low level of Agreement sheds doubts on the grades’ validity. The worst case occurs when they disagree on every dish.</p><p>The Kappa score’s concept is to benchmark the level of Agreement, i.e., identical grades assigned by Chef A and Chef B, which were obtained versus the Chance Agreement. The Chance Agreement is the level of Agreement we would get if the two chefs were evaluating the dishes entirely randomly. The number of elements to classify in this example is 74.</p><p>For instance, Chef A and B could issue the following grades for the first 10 dishes (among 74 evaluated):</p><p>DishChef AChef B1ExquisiteNo2NoNo3NoMaybe4MaybeMaybe5ExquisiteExquisite6MaybeExquisite7NoExquisite8MaybeNo9NoExquisite10ExquisiteExquisite</p><p>And proceed with the remainder of the 74 dishes up for culinary competition. We can compile the results into a matrix that presents the number of recipes for which the chefs agreed (diagonal) and the number of dishes of disagreement, as shown below.</p><p>Chef A Chef B ExquisiteNoMaybeExquisite 762No 2325Maybe 1316</p><p>So now, how do we determine and compare the Agreement and Chance Agreement? Let’s start with the Chance Agreement.</p><p>The chance that Chef A picks randomly Exquisite is the number of times Chef A chose Exquisite over the total number of dishes:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_5cNb2sGBk6hwHDw.png" /></figure><p>Applying the same approach, we can calculate ChanceB(Exquisite), i.e., 0.28, also as the number of Exquisite selected by B divided by the total number of dishes.</p><p>Both Chef A and Chef B pick exquisite simultaneously by chance with a probability, which is the product of ChanceA(Exquisite) and ChanceB(Exquisite), i.e., 0.04. Following this method, we can calculate Chance(No), i.e., 0.28, and Chance(Maybe), i.e., 0.08. The overall chance agreement is the sum of these three chances: Chance(Exquisite), Chance(No), and Chance(Maybe), i.e., 0.4.</p><p>The Agreement is the sum of all the grades the chefs agreed on (the diagonal in the previous matrix) divided by the total number of occurrences, here it is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/921/0*yMCklcWbr5oAGoRK.png" /></figure><p>Once we have computed both the Agreement and the Chance Agreement, the Kappa score is defined as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*o0ERPGEP0x_QYSQn.png" /></figure><p>The Kappa score for our chefs rating of the 74 dishes is 0.56. In the worst case, the Agreement is identical to the Chance Agreement.</p><p>The three options: Exquisite, No, and Maybe can be interpreted as three classes. And the two chefs, A and B, can be transposed in the machine learning world into ‘Predicted’ and ‘Actual.’ That’s how the Kappa score is transcribed from the world of psychology to Machine Learning and sorting algorithms. And that’s how computer science has adopted a concept stemming from psychology.</p><p>When applied to multiclass classification, the Kappa score reveals how well the model has predicted data assignments in different classes compared to a random class assignment. If the predictions have well identified the categories, the Kappa Score gets close to 1. On the contrary, if the sorting algorithm does not determine the classes well, the Kappa Score becomes close to 0 in the worst case.</p><h3>Multiclass and sorting algorithms in a nutshell</h3><p>We have reviewed the metrics considered for binary classification and applied them to multiclass classification regardless of the number of elements. These metrics stay readable and interpretable, even in the context of an expanded number of classes.</p><p>We have also discovered a new metric, borrowed from the realm of psychology, the Kappa score. This metric correlates the level of True Positives obtained by the algorithm under scrutiny, versus a random procedure.</p><p>The metrics available to estimate the performance of sorting algorithms are varied. Some metrics are more appropriate when the classes are well balanced; others emphasize the algorithm’s effectiveness with unbalanced classes. In a single figure, the Kappa score shows how well the model performs compared to a random algorithm.</p><h3>References</h3><p><a href="https://towardsdatascience.com/multi-class-metrics-made-simple-part-ii-the-f1-score-ebe8b2c2ca1">https://towardsdatascience.com/multi-class-metrics-made-simple-part-ii-the-f1-score-ebe8b2c2ca1</a></p><p><a href="https://towardsdatascience.com/multi-class-metrics-made-simple-part-i-precision-and-recall-9250280bddc2">https://towardsdatascience.com/multi-class-metrics-made-simple-part-i-precision-and-recall-9250280bddc2</a></p><p><a href="https://towardsdatascience.com/multi-class-metrics-made-simple-the-kappa-score-aka-cohens-kappa-coefficient-bdea137af09c">https://towardsdatascience.com/multi-class-metrics-made-simple-the-kappa-score-aka-cohens-kappa-coefficient-bdea137af09c</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2fa8f76e37e7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/multiclass-classification-sorting-algorithms-2fa8f76e37e7">Multiclass Classification: Sorting Algorithms</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Regression Algorithms: Which Machine Learning Metrics?]]></title>
            <link>https://medium.com/mydatamodels/regression-algorithms-which-machine-learning-metrics-9b8ee9f060f8?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/9b8ee9f060f8</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[metrics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Tue, 03 Nov 2020 10:33:06 GMT</pubDate>
            <atom:updated>2020-11-03T10:32:47.771Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QfEkDsUTaY_5RFNR.jpg" /></figure><p>Can Machine Learning predict the size of a person based on his/her age and weight? Can Machine Learning compute the price of a house based on its surface, location, and number of rooms? Can Machine Learning determine the revenue of a company based on its marketing budget? These are all typical examples of regression problems.</p><h3>Metrics for Regression Algorithms</h3><p>Last time we wrote about <a href="https://www.mydatamodels.com/how-good-is-your-machine-learning-algorithm">binary classification</a> and the corresponding Machine Learning Metrics. We explained that Binary Classification consists of assigning an individual to one of two classes. An illustration is a medical diagnosis for COVID-19 infection. A person is either infected or not.</p><p>Regression, in contrast to binary classification, provides an estimate of a continuous value. For instance, what is the forecasted height of a male subject aged 52 who weighs 80 kilos?</p><p>Another instance of a regression problem could be to evaluate the sales (in units) of a new breakfast cereal based on the advertising budgets allocated for Facebook, Instagram, Twitter, TV, and radio. We might also need to understand each media’s contribution to sales to optimize the marketing funds further. The underlying link between the age, the weight of a person, and his size is a common regression problem. Similarly, the dependency between a marketing budget and the resulting sales is a regression case.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*O5S-iNyQfaZfP8TF.png" /></figure><p>Each point on the graph is the cereals sales at a specific time. The data points vary as the marketing budget increases. Still, there is (in this case) a linear relationship between the two, where the revenue increases with the marketing budget. The idea is that the dependency between two variables (in our example, the marketing budget and the sales) can be modelled through a graph.</p><p>This graph might be linear, or it might be Gaussian or polynomial. This graph allows us to extrapolate what the sales will be, based on the marketing budget planned. The best regression algorithm is the one that best fits the chart with the actual data points.</p><p>Let’s consider an existing house.If the data points were the 3D geographical coordinates of numerous house walls, the perfect regression model would be the house’s original floor plan. The more critical the errors, the less accurate the graph, and hence the regression model. How can errors be measured in the case of regression algorithms?</p><h4>Machine Learning Metrics: Cost Function</h4><p>When building a regression model, we are attempting to reduce the error an algorithm does. To do that, we select a function to measure the error, also called cost function.</p><p>The Machine Learning Metrics we have seen so far for classification, such as accuracy, do not apply to regression. Instead, we use metrics designed for analyzing continuous values. The different Machine Learning Metrics used to evaluate the results of the prediction are :</p><ul><li>Mean Absolute Error (MAE),</li><li>Mean Absolute Percentage Error (MAPE),</li><li>Mean Squared Error (MSE),</li><li>Root-Mean-Squared-Error (RMSE),</li><li>Maximum Error (ME),</li><li>R² or Coefficient of Determination.</li></ul><h4>Machine Learning Metrics: Mean Absolute Error</h4><p>MAE is the average of the absolute differences between the actual value and the model’s predicted value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/685/0*nUXaeZX-jaQakGQ8.png" /></figure><p>The bigger the MAE, the more critical the error is. The MAE unit is the same as the predicted variable unit, i.e., a distance is estimated in km, a weight in kilograms.</p><p>Therefore, the MAE cannot compare regression models’ performance for distinct categories of data. It is robust to outliers, i.e., extreme values. Hence, it is not suitable for applications where you want to pay more attention to these outliers.</p><h4>Machine Learning Metrics: Mean Absolute Percentage Error</h4><p>MAPE or Mean Absolute Percentage Error is the average absolute difference between the actual value and the value predicted by the model divided by the real value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/729/0*mL2ufqVwm_0yjR4h.png" /></figure><p>Its usage is comparable to the MAE, only, since it is a percentage, it allows for comparison between regression models designed for diverse categories of data. It does not give a specific focus to outliers. However, in some cases, we want to use a cost function that emphasizes outliers.</p><h4>Machine Learning Metrics: Mean Squared Error and Root Mean Squared Error</h4><p>MSE or Mean Squared Error is one of the most popular metrics for regression algorithms. It is merely the average of the real value’s squared difference with the regression model’s predicted value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/685/0*gdBHnFNXLyUjfc6k.png" /></figure><p>As it squares the differences, it is harder on outliers, leading to over-estimating how bad the model is. Its unit is the square of the variable’s unit.</p><p>The RMSE or Root Mean Squared Error is the average root-squared difference between the real value and the predicted value. Its use is similar to the MSE.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/685/0*aVGFZM9udpZAEBQQ.png" /></figure><h4>Machine Learning Metrics: Maximum Error</h4><p>ME or Maximum Error is the absolute value of the most significant difference between a predicted variable and its real value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/855/0*E2cvm7nN-_Lv1Exs.png" /></figure><p>It is interesting to spot out rapidly how well the model integrates outliers. Typically, if the Maximum Error is much bigger than the RMSE, it might mean that the model has not correctly predicted outliers.</p><h4>Machine Learning Metrics: R² or Coefficient of Determination</h4><p>R² or Coefficient of Determination is a prevalent metric. R² uses two mean squared error calculations. While the former is the mean square of each real value versus the average of observations, the latter is the mean squared error of the actual value versus the predicted one.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/556/0*ECEJMmZCMEOfSJUn.png" /></figure><p>R² is the one complement of the ratio between these two MSE. R² score ranges from -∞ to 1. The closest to 1 the R², the better the regression model is. If R² is equal to 0, the model is not performing better than a random model.</p><p>If R² is negative, the regression model is erroneous. Therefore this last Machine Learning Metric is an excellent tool to evaluate the efficiency of a regression model.</p><h4>Machine Learning Metrics for Regression In a Nutshell</h4><p>Let’s consider a perfectly well-centered Gaussian model with very few outliers, all close to the average. In this case, the RMSE is an excellent metric to use. We have seen various metrics to evaluate the performance of different regression algorithms. Depending on the situation, some metrics might be more relevant than others.</p><p>In all cases, the goal is to get a fair estimate of the distance between the predicted model and the actual values. Whether the MAE, MAPE, MSR, RMSE, ME, or R² is used, the critical point is to have a subtle understanding of these metrics to use them adequately.</p><p>Stay tuned for our next article about multi-class metrics !</p><h3>References</h3><p><a href="https://towardsdatascience.com/introduction-to-machine-learning-algorithms-linear-regression-14c4e325882a">https://towardsdatascience.com/introduction-to-machine-learning-algorithms-linear-regression-14c4e325882a</a></p><p><a href="http://www.cs.cmu.edu/afs/andrew/course/15/381-f08/www/lectures/regression.pdf">http://www.cs.cmu.edu/afs/andrew/course/15/381-f08/www/lectures/regression.pdf</a></p><p><a href="https://www.unemyr.com/simple-linear-regression-ai/">https://www.unemyr.com/simple-linear-regression-ai/</a></p><p><a href="https://medium.com/simple-ai/linear-regression-intro-to-machine-learning-6-6e320dbdaf06">https://medium.com/simple-ai/linear-regression-intro-to-machine-learning-6-6e320dbdaf06</a></p><p><a href="http://faculty.marshall.usc.edu/gareth-james/">http://faculty.marshall.usc.edu/gareth-james/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9b8ee9f060f8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/regression-algorithms-which-machine-learning-metrics-9b8ee9f060f8">Regression Algorithms: Which Machine Learning Metrics?</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deaths And COVID AI]]></title>
            <link>https://medium.com/mydatamodels/deaths-and-covid-ai-6d2d5da23187?source=rss----7901bf652483---4</link>
            <guid isPermaLink="false">https://medium.com/p/6d2d5da23187</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[covid19]]></category>
            <dc:creator><![CDATA[MyDataModels]]></dc:creator>
            <pubDate>Thu, 29 Oct 2020 11:10:20 GMT</pubDate>
            <atom:updated>2020-10-29T11:09:50.960Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hzmjAfSzlEDZgO_0.jpg" /></figure><p>Artificial Intelligence applied to COVID, i.e., COVID AI, is key to fighting the deadly virus. Over the past ten months, the virus has taken more lives than H.I.V., malaria, influenza, and cholera over the same period. And the COVID-19 pandemic continues to propagate around the globe.</p><h3>Fighting deaths: entering Clinical Trials</h3><p>Since outbreaks started at the end of 2019, the coronavirus has ripped through country after country. According to Johns Hopkins University data, it has been sickening more than 34 million people worldwide and caused over 1.13 million mortality. A severe toll gathered from official counts, yet one that notably underestimates how many have really perished. COVID-19 is the underlying cause of deaths attributed to other diseases.</p><p>The high death rate reported arrives ten months after the first cases in Wuhan, China. Coronavirus may already have beaten tuberculosis and hepatitis as the world’s deadliest infectious disease in the general population. And the mortality statistics are growing fast. Still, it is far behind the overall deaths attributed to H.I.V. with 32 million since its first occurrences.</p><p>In comparison, it is nowhere near the lethal pandemic of the 1918 Spanish flu that claimed 50 million lives across countries. The coronavirus’s survival rate and mortality toll make it unprecedented and deadly among modern epidemics. In contrast, the 2009 H1N1 flu demanded 18,500 lives. The threat of these infectious diseases forces countries to focus on control and prevention.</p><p>A few countries such as China, Germany, South Korea, and New Zealand, have slowed the pandemic enough to limit infections and deaths while still reopening businesses and schools. It requires a combination of well-coordinated government actions that may be difficult to achieve in some countries: wide-scale testing, contact tracing, quarantining, social distancing, mask-wearing, developing a clear and consistent strategy, and being ready to lock things down in a rush when difficulties occur.</p><p>But the countries which have managed to handle this crisis are scarce. Many look with hope at a potential vaccine to reduce the mortality rate attributed to COVID-19. Yet, vaccine trials enter their final phases toward the end of 2020. And there is no such thing as a 100% successful vaccine.</p><p>Furthermore, not all citizens will get vaccinated even though it might cause self-harm. Therefore, even a successful vaccine may not replace the disease control and prevention methods. Besides, it remains unclear how the virus mutates; hence it is impossible to predict how long a possible vaccine might work.</p><h3>COVID AI: what makes people more prone to death?</h3><p>As a data scientist team, we have tried to get some insights from data regarding what caused people to die from COVID-19. The data we have chosen is a Mexican database. It is made publicly available by the Mexican government.</p><p>It is cited in various research studies and papers. One of these studies is an article by the medical researcher Omar Yaxmehen Bello-Chavolla and his team.</p><p>The records in this database are inputs from patients checking in Mexican hospitals with various conditions: cardiovascular diseases, respiratory diseases, kidney diseases, influenza, and pneumonia. It includes the gender, the age-groups, the comorbidities. We have already analyzed this database from the perspective of <a href="https://www.mydatamodels.com/coronavirus-predictions-and-covid-19-infections/">COVID-19 infections</a>, entry in <a href="http://coronavirus-disease-and-intensive-care-unit/">I.C.U.</a> and <a href="https://www.mydatamodels.com/covid-19-intubation-to-fight-respiratory-problems/">intubation</a>.</p><p>We have run our Machine Learning tool TADA over this database to understand the causes of morbidity and mortality. The question we asked the tool was, “given a COVID-19 patient, what risk factors are likely to become causes of death?”. TADA created numerous machine learning models and extracted the most impacting elements from them. Globally, the rate of death among COVID-19 patients is less than 5%.</p><p>The average accuracy of the resulting COVID AI models was 86%, with an 87% sensitivity. It means that the model reasonably predicts the five percent of death. The false negatives, i.e., people expected to live who died, represented one percent. These figures account for excellent models.</p><p>The main criteria used by the ai models to predict the death of a COVID-19 patient were:</p><ul><li>whether the patient was intubated,</li><li>had pneumonia,</li><li>belonged to an age-group over 65,</li><li>and was a contact case.</li></ul><p>Pneumonia was an overwhelming criterion and, consequently, considered one of the leading causes of death. We decided to explore what the COVID AI model had to say when pneumonia was discarded. So we reran TADA without pneumonia, the resulting models showed an accuracy of 80% and a similar sensitivity.</p><p>The impacting criteria stayed identical, except for pneumonia, of course: intubation, age, contact case. However, a new criterion appeared in this new study: obesity. Chronic obstructive pulmonary disease (COPD) does not appear as a leading cause of death in this study, nor do other pulmonary diseases, diabetes, or coronary diseases.</p><h3>The Fight Against COVID-19 Pandemic with Artificial Intelligence</h3><p>This article is the last of a series of four about COVID-19 and Machine Learning. We have shown that it is possible to propose strong hypotheses for further investigation by medical research with Machine Learning models. Age appears to be a significant aggravating criterion for all the stages we have analyzed: contamination, entry in emergency care, intubation, and death.</p><p>For contamination and entry in I.C.U., diabetes was a criterion which surprised us. In this last article, we were astonished to see that obesity was a leading cause of death in COVID-19 patients.</p><p>We do not claim to invent COVID treatment for COVID-19 patients. We claim to fight COVID-19 by helping medical researchers be quicker with the data they gather.</p><h4>References</h4><p><a href="https://www.cdc.gov/coronavirus/2019-ncov/covid-data/investigations-discovery/hospitalization-death-by-age.html">https://www.cdc.gov/coronavirus/2019-ncov/covid-data/investigations-discovery/hospitalization-death-by-age.html</a><br><a href="https://www.worldometers.info/coronavirus/">https://www.worldometers.info/coronavirus/</a><br><a href="https://edition.cnn.com/2020/09/30/us/app-state-student-covid-19-death-trnd/index.html">https://edition.cnn.com/2020/09/30/us/app-state-student-covid-19-death-trnd/index.html</a><br><a href="https://www.npr.org/sections/goatsandsoda/2020/09/28/916984072/covid-19-deaths-top-1-million-how-these-5-countries-are-driving-the-pandemic?t=1601622733380">https://www.npr.org/sections/goatsandsoda/2020/09/28/916984072/covid-19-deaths-top-1-million-how-these-5-countries-are-driving-the-pandemic?t=1601622733380</a><br><a href="https://www.nytimes.com/2020/09/28/world/covid-1-million-deaths.html">https://www.nytimes.com/2020/09/28/world/covid-1-million-deaths.html</a><br><a href="https://www.cbsnews.com/news/coronavirus-deaths-1-million-worldwide/">https://www.cbsnews.com/news/coronavirus-deaths-1-million-worldwide/</a><br><a href="https://www.unaids.org/en/resources/fact-sheet">https://www.unaids.org/en/resources/fact-sheet</a><br><a href="https://www.nytimes.com/2020/05/06/world/europe/germany-merkel-coronavirus-reopening.html">https://www.nytimes.com/2020/05/06/world/europe/germany-merkel-coronavirus-reopening.html</a><br><a href="https://www.nytimes.com/2020/03/23/world/asia/coronavirus-south-korea-flatten-curve.html">https://www.nytimes.com/2020/03/23/world/asia/coronavirus-south-korea-flatten-curve.html</a><br><a href="https://www.nytimes.com/2020/08/13/world/asia/new-zealand-coronavirus-lockdown-elimination.html">https://www.nytimes.com/2020/08/13/world/asia/new-zealand-coronavirus-lockdown-elimination.html</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6d2d5da23187" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mydatamodels/deaths-and-covid-ai-6d2d5da23187">Deaths And COVID AI</a> was originally published in <a href="https://medium.com/mydatamodels">MyDataModels</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>