<?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 Adam Słucki on Medium]]></title>
        <description><![CDATA[Stories by Adam Słucki on Medium]]></description>
        <link>https://medium.com/@adam.slucki?source=rss-41c110f2afb5------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*0W1vNVi6uwHTe_Hm.jpg</url>
            <title>Stories by Adam Słucki on Medium</title>
            <link>https://medium.com/@adam.slucki?source=rss-41c110f2afb5------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 21:06:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@adam.slucki/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[5 AI trends in the times of COVID]]></title>
            <link>https://medium.com/tooploox-ai/5-ai-trends-in-the-times-of-covid-d7ebb24f4f94?source=rss-41c110f2afb5------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7ebb24f4f94</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[covid19]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Adam Słucki]]></dc:creator>
            <pubDate>Fri, 22 May 2020 07:23:08 GMT</pubDate>
            <atom:updated>2020-05-22T07:23:08.705Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FL0ammjGm1cKxVbNuN1AxQ.jpeg" /></figure><p>Current situation in the world forced Machine Learning scientists and ML-oriented companies to laser focus on AI solutions that can help to fight the coronavirus. We selected 5 trends in AI that are especially beneficial in current difficult times.</p><h3>Time-series prediction</h3><p>How long will coronavirus last in my country? How many people are going to die? How long should we keep the lockdown? The answer to that questions may seem to be a bit of a fortune telling, but day by day we collect more and more data, many countries apply different strategies to deal with the coronavirus and we have some overview of how political decisions influence the curves that show the total diagnosed cases and deaths. Researchers from <a href="https://alphamoon.ai/">Alphamoon</a> tackled the problem by enriching the epidemical SEIR model with Bayesian modeling. In their blog, they provide models for three scenarios for Poland: releasing the lockdown after Easter, keeping the lockdown and keeping the lockdown &amp; increasing the number of tested people. For the detailed results and discussion, you can visit their <a href="https://alphamoon.ai/blog/2020/03/24/how-to-predict-the-future-of-sars-cov-2-pandemic/">blog</a>.</p><h3><strong>Image classification and segmentation</strong></h3><p>One of the main bottlenecks in current coronavirus tests is a long waiting time for the diagnosis. To reduce the delay between taking a sample and the test result scientists are searching for alternative ways of testing. One possible way to do that is to detect COVID-19 based on Computerized Tomography (CT) scans. <a href="https://www.medrxiv.org/content/10.1101/2020.02.14.20023028v4.full.pdf">Studies of Wang et. al. </a>show that this diagnostic problem can be solved by deep classification models. Authors of the proposed solution claim that with this approach they can achieve detection accuracy at 89.5% with balanced specificity of 0.88 and sensitivity of 0.87.</p><h3><strong>Machine learning models for drug discovery</strong></h3><p>The process of discovering the new drugs is very time consuming and taking a drug from research to market is estimated to cost on average 2.6 billion dollars, and last more than 10 years (<a href="https://towardsdatascience.com/unlocking-drug-discovery-through-machine-learning-part-1-8b2a64333e07"><em>see here for details</em></a><em>)</em>. The early stage of the drug discovery process can be significantly accelerated with machine learning solutions like deep generative models. The main challenge for the models in this field is to generate novel and valid molecules. Companies like <a href="https://www.protocol.com/ai-machine-learning-coronavirus-covid19">Insilico</a> and <a href="https://deepmind.com/blog/article/AlphaFold-Using-AI-for-scientific-discovery">AlphaFold</a> are working on the problem and they have some initial successes in that field.</p><h3><strong>Social Network Analysis</strong></h3><p>Social media plays a crucial role in spreading information. Decentralisation of this medium has a lot of positive aspects but at the same time it generates at least two problems. Firstly, meaningful information can be lost in a flood of news generated by more popular actors on the social media scene. Secondly, fake news and clickbaits can create a false and dangerous picture of the situation that in turn may lead to destructive behaviours either like or for example (not both) ignoring recommendations of health organisations.</p><p>To prevent this problem AI systems can be used to analyse news appearing on social media in real time and assess their credibility as well as detect so-called fake news. <a href="http://www.fakenewschallenge.org/">Fake news detection</a> is a growing branch of Natural Language Processing.</p><h3><strong>Object detection on videos</strong></h3><p>Object detection itself is one of the most actively researched problems in Computer Vision. Applying the same problem to videos instead of static images poses greater challenges but at the same time gives more possibilities for real life applications, including applications aimed to help with the current epidemic. In China robots equipped with cameras were checking if pedestrians were wearing protective masks and warning them if they didn’t. Real time video analysis can be used also to monitor health conditions of patients. It could be especially useful in a case of shortage of specialized medical equipment. There are even ideas to apply such technology so it can assist in proper hand washing to minimize the risk of infection. Tooploox is also working on AI solutions to deal with coronavirus. We’ve created a visual AI system that is able to detect gatherings on monitoring footages. The demonstration video is available <a href="https://www.youtube.com/watch?v=BIpapSWRsBE">here</a>.</p><p>The post was prepared collaboratively by Maciej Zięba and Adam Słucki.</p><p>If you want to read more about AI, check out our site at Tooploox: <a href="https://www.tooploox.com/ai">https://www.tooploox.com/ai</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7ebb24f4f94" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/5-ai-trends-in-the-times-of-covid-d7ebb24f4f94">5 AI trends in the times of COVID</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The effect of isolation — are lockdowns helpful?]]></title>
            <link>https://medium.com/tooploox-ai/the-effect-of-isolation-are-lockdowns-helpful-a87ece5ebf96?source=rss-41c110f2afb5------2</link>
            <guid isPermaLink="false">https://medium.com/p/a87ece5ebf96</guid>
            <category><![CDATA[epidemic]]></category>
            <category><![CDATA[covid19]]></category>
            <category><![CDATA[multi-agent-systems]]></category>
            <dc:creator><![CDATA[Adam Słucki]]></dc:creator>
            <pubDate>Wed, 13 May 2020 13:54:05 GMT</pubDate>
            <atom:updated>2020-05-15T08:27:54.077Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Kspsy6HQHG-9Rgtf6XERvA.jpeg" /></figure><h3><strong>The effect of isolation — are lockdowns helpful?</strong></h3><p>We all see that the regulations implemented in many countries due to the COVID-19 pandemic have tremendous effects on the economy. As much as the world seems to be digitized it is still ill prepared for the majority of the population to be locked in their homes. Among many frustrations, it raises the question if staying at home is truly of any help in this situation? But how can we answer this? Well, one approach is to simulate reality to test various scenarios.</p><p>In this post we will introduce you to the topic of multiagent based simulations and show you the results of the simulation our team performed here at Tooploox.</p><h3>How to simulate a complex world</h3><p>Thinking of simulations, often something extremely realistic comes to mind, like a flight simulation, which wouldn’t be at all useful without a given level of attention to detail. But when we try to simulate phenomena on a larger scale rather than focus on individuals, then simplification is the key.</p><p>Let’s look at some examples. Below is a detailed simulation of respiratory droplet dispersion to visualize what happens when someone coughs indoors. You can read about it here: <a href="https://www.nytimes.com/interactive/2020/04/14/science/coronavirus-transmission-cough-6-feet-ar-ul.html">post on www.nytimes.com</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/619/0*eC6AHbvdZ2kdZuN2" /><figcaption>Source: <a href="http://www.nytimes.com/">www.nytimes.com</a></figcaption></figure><p>This is a simplified simulation where the physical mechanism of a contagion is using the assumption: <strong>the closer you get to an infected person the greater the chance you’ll be infected</strong>.<strong> </strong>This allows you to isolate only those variables that are most important to you. It also reduces the required computing power, especially when the intention is to simulate phenomena occurring among large groups of subjects.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*dD8eqTQdAuS9laar" /></figure><p>So to simplify the complexity related to the epidemic we’ll need:</p><ul><li><strong>Environment:</strong> this is a space where all the action will occur. It also encapsulates assumptions about the real world. In our case, these will be how many persons are infected at the beginning of the simulation, how easy it is to get infected and how eager people are to leave their homes.</li><li><strong>Agents: </strong>autonomous objects reflecting the subject of our study, in this case living persons that can become infected when in contact with the virus.</li><li><strong>Interactions: </strong>the key aspect of the simulation is that we set relatively simple rules of how agents can interact with each other and the environment, and this often leads to an <a href="https://en.wikipedia.org/wiki/Emergence">emergence</a>, allowing us to create truly complex systems. A very prominent example of the emergent system is <a href="https://en.wikipedia.org/wiki/Langton%27s_ant">Langton’s ant</a>.</li></ul><h3>Our simulated community</h3><p>To perform our simulations we defined multiple environments. One of them was <strong>a small district with a single store</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*OiT38--3ST3EpA3h" /><figcaption>Fragment of the simulated environment.</figcaption></figure><p>So in our <strong>environment </strong>we have 400 agents in total and homes of different sizes that can fit from one to nine agents. Agents are assigned to homes randomly and there are about 130 homes, so we can expect that on average there will be 3 agents in every home — the most common setup in developed countries. Each agent can go to the store for a given amount of time, but <strong>there is a limit to the number of agents that can be in the store at the same time</strong>. And this limit is what interests us the most — we want to test how it affects the number of infected agents.</p><p><strong>The agents</strong> can have one of three states: healthy, infected and recovered. They are also somehow intelligent. They try to keep distance from each other and walk inside the store with a specific goal in mind so their moves are not random. We achieved that with a customised variation of the A* algorithm.</p><p>The mechanism of contagion is simple. When a healthy agent is in close range to an infected agent, it can catch the virus with a given probability. The closer the distance, the higher the probability. Once the agent is infected it stays contagious for 140 ticks of the simulation and then becomes recovered and immune. It follows the SIR compartmental model of infectious diseases like COVID-19, but in a multi-agent simulation we also have a spatial aspect which makes it more realistic. These are elements of the SIR model:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/849/0*nQKIMit7hqXa2LQt" /></figure><p>If you’re interested in the technicalities, we used the <a href="https://github.com/projectmesa/mesa">MESA framework</a> to build the simulations. We also slightly modified some visual aspects in the original library.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fy84FPdJSYKI%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dy84FPdJSYKI&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fy84FPdJSYKI%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/be885240b75a8ac6296fcc08e2d63d1c/href">https://medium.com/media/be885240b75a8ac6296fcc08e2d63d1c/href</a></iframe><h3>Simulation — Does it matter if more people can be in the store at the same time?</h3><p>In our simulation, agents staying at home can be infected only by family members. <strong>The only way of spreading the disease through the whole community is by meeting infected agents at the store</strong>. We define two only slightly different scenarios to check how they will affect the percentage of the infected population in 14 days (140 steps of the simulation):</p><p><strong>Scenario I:</strong> 5 agents are allowed to be in the store at the same time.</p><p><strong>Scenario II: </strong>7 agents are allowed to be in the store at the same time.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FM2FDxOqInyY%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DM2FDxOqInyY&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FM2FDxOqInyY%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/8e7bf9db556873f96440c960209594ac/href">https://medium.com/media/8e7bf9db556873f96440c960209594ac/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FCs6II-2fU2o%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DCs6II-2fU2o&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FCs6II-2fU2o%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/1879be1b9ca3105c22295e7c29a66bf1/href">https://medium.com/media/1879be1b9ca3105c22295e7c29a66bf1/href</a></iframe><h4>Comparing the results</h4><p>As we simplify things we must introduce some randomness to our simulations. It alleviates missing details we couldn’t or didn’t want to deal with. To do so we have to run the simulation with each scenario multiple times. We did 15 runs for each scenario and collected the maximum percentage of the infected population in 14 days. Here are the results:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/466/1*T6A65teqvq2jmwCbbp65Ag.png" /><figcaption>The highest percentage of infected agents in each run in Scenario I and Scenario II.</figcaption></figure><p>The<strong> mean value </strong>of observations<strong> </strong>for <strong>scenario I </strong>is<strong> 9.93 </strong>and<strong> </strong>for <strong>scenario II</strong> is<strong> 12.07</strong>. To check if the difference between the means is not just a result of a random variation, we have to perform a statistical test called Student’s t-test. The test gives a <a href="https://en.wikipedia.org/wiki/P-value">p-value</a> equal 0.0038. This means that if there was really no difference between the mean values obtained in each scenario, such a result, or even more extreme one, would be observed in 3.8 cases (i.e. running simulations 15 times and comparing the means) out of 1000. This is very unlikely. Also, produced p-value is much smaller than a commonly accepted significance level threshold value of 0.05 so we conclude that the difference is statistically significant, i.e. that the effect is real.</p><h3>Simulation — When is a good time to come out of lockdown?</h3><p>At this point we are surely all <strong>tired of being kept on lockdown</strong> and look forward to going on freely with our lives — going out with friends, attending bigger cultural/sports events and traveling. The good news is that we’re seeing a steep <a href="https://www.endcoronavirus.org/countries">decline in the number of new cases in an increasing number of countries</a>. But opening things up too soon might cause a surge in new cases and a second wave of the epidemic.</p><p>To find out when is a good moment to open the economy in our simulated environment, we allowed agents to meet in bigger crowds (up to 20 agents in the public space, which is quite crowded for our little environment) and verified the following initial conditions:</p><p><strong>Scenario I:</strong> public gatherings are allowed when <strong>10%</strong> of the population has already recovered.</p><p><strong>Scenario II:</strong> public gatherings are allowed when <strong>15%</strong> of the population has already recovered.</p><p><strong>Scenario III: </strong>public gatherings are allowed when <strong>20%</strong> of the population has already recovered.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJfiCNpu71Po%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJfiCNpu71Po&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FJfiCNpu71Po%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/2d10e6a5ae0720713dbce9666ff6f764/href">https://medium.com/media/2d10e6a5ae0720713dbce9666ff6f764/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FgrF4Pxv419E%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DgrF4Pxv419E&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FgrF4Pxv419E%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6414dc16b910484cec5815f02fbc0e5a/href">https://medium.com/media/6414dc16b910484cec5815f02fbc0e5a/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FPQSkEjLcrHE%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DPQSkEjLcrHE&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FPQSkEjLcrHE%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/5298fafe0192a1e2c657676dfee2daf8/href">https://medium.com/media/5298fafe0192a1e2c657676dfee2daf8/href</a></iframe><h4>Comparing the results</h4><p>The number of initially infected agents is the same for all scenarios <strong>— 3%</strong>. We only allowed agents to meet in one public space (no visiting neighbours was allowed). Similarly to the previous use case, we ran the simulation 15 times and calculated the average value of infected agents.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/502/1*thetCPlNd-Gfz_hRrQX-qA.png" /><figcaption>The highest percentage of infected agents in each run in Scenario I, II and III.</figcaption></figure><p>What we observe from the simulations is that even a small difference in the initial population of recovered agents causes a significant difference when it comes to the maximum percentage of infected agents. Opening the store for up to 20 agents (and allowing the space to get crowded) with only 10% of recovered agents caused a second wave epidemic with almost a third of the population getting sick.</p><h3>Conclusion</h3><p>Simulations show that limitations regarding how many people can enter indoor facilities at the same time make sense and even seemingly slight changes can have a visible impact on the larger scale.</p><p>Although a few percentage points may look like nothing, we must remember that nominal values in this case really do matter. Even 1% of a larger population means thousands of people which can be catastrophic considering a shortage of a medical staff and equipment. We’ve seen the effects of the outbreak in Italy, and other countries, like <a href="https://www.bbc.com/news/av/world-asia-52400084/coronavirus-tokyo-hospitals-trying-to-stay-ahead">Japan, are struggling too</a>.</p><p>In many countries we observe a transition from the “hammer” phase to the “dance” phase, but nobody knows the correct strategy to come out of lockdown while keeping the number of cases under control. The <a href="https://www.nytimes.com/2020/05/07/world/europe/coronavirus-reopening-costs.html?campaign_id=51&amp;emc=edit_MBE_p_20200508&amp;instance_id=18312&amp;nl=morning-briefing&amp;regi_id=120882415&amp;section=topNews&amp;segment_id=26884&amp;te=1&amp;user_id=757862cf10c3ea77f9821ee95653ee4b">first wave of reopenings in Europe and Asia</a> do not seem to follow a unified path and resembles a trial-and-error process. While everybody is eager to go outside, opening the economy with a small percentage of recovered population and a small percentage of infections may cause a second wave of infection, just as we have observed in the simulated environment.</p><p><strong>It’s better to stay safe — to stay at home.</strong></p><p>The project was created collectively by Ivona Tautkute, Konrad Czarnota and Adam Słucki.</p><p>If you want to read more about AI, check out our site at Tooploox: <a href="https://www.tooploox.com/ai">https://www.tooploox.com/ai</a></p><p><strong>Sources:</strong></p><ol><li>Data for input parameters: <a href="http://weekly.chinacdc.cn/en/article/id/e53946e2-c6c4-41e9-9a9b-fea8db1a8f51">Vital Surveillances: The Epidemiological Characteristics of an Outbreak of 2019 Novel Coronavirus Diseases (COVID-19) — China, 2020</a></li><li>Old but easy to grasp explanation of multi agent simulations: J. M. Epstein, R. Axtell. Growing Artificial Societies: Social Science From the Bottom Up. The MIT Press, 1996.</li><li>More technical article regarding multi agent simulations in epidemiology: Roche B, Guégan JF, Bousquet F. Multi-agent systems in epidemiology: a first step for computational biology in the study of vector-borne disease transmission. BMC Bioinformatics, 2008.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a87ece5ebf96" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/the-effect-of-isolation-are-lockdowns-helpful-a87ece5ebf96">The effect of isolation — are lockdowns helpful?</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Top 3 Projects from the “Hack the Crisis” Hackathon]]></title>
            <link>https://medium.com/tooploox-ai/top-3-projects-from-the-hack-the-crisis-hackathon-3572e07a4566?source=rss-41c110f2afb5------2</link>
            <guid isPermaLink="false">https://medium.com/p/3572e07a4566</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[hackathons]]></category>
            <category><![CDATA[coronavirus]]></category>
            <dc:creator><![CDATA[Adam Słucki]]></dc:creator>
            <pubDate>Thu, 09 Apr 2020 11:05:38 GMT</pubDate>
            <atom:updated>2020-04-09T11:10:35.180Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*05wToM6kSBUP6iN_BaldAA.jpeg" /></figure><p>In March (17–22.03.2020) a hackathon focused on solving problems related to the COVID-19 virus pandemic took place. There were around 200 ideas submitted and 10 of them were awarded a prize. <strong>We’ve selected the 3 we liked the most.</strong></p><h3>Detecting COVID-19 on X-rays with deep learning</h3><p>We would rather call the project “Diagnosing COVID-19 based on X-ray images using deep learning” to be more precise. Anyways, one of the problems with COVID-19 is that infected people may show no symptoms of the disease. As resources for medical examination are very limited, such people have very little chance to be diagnosed and properly quarantined.</p><p>To solve this problem by ensuring easier access to examination, the team proposes an AI model that uses the x-ray images of lungs to diagnose if they are infected or not. By using an automated system, both the time and personnel required for analysing x-ray images could be notably reduced. It would allow for more people to undergo such examinations in a shorter time.</p><p>Unfortunately this idea is difficult to implement, at least at this moment, due to the lack of data required to train and evaluate such AI models. X-ray images, publicly available at the time of training, were presenting patients in serious condition where the infection was obvious. It could be the case that early stage development diagnosis based on the visual inspection of x-ray images of the lungs is inaccurate or less effective than other methods.</p><p>Still, we believe that if the required data would be available, then work on this approach should be continued. If extensive evaluation would show that the solution provides an efficient tool for diagnosis, it would not only be helpful during the epidemic but also in the future in every case where a similar approach could be deployed.</p><h3>Medy — be rewarded for prevention ✨</h3><p>Disinformation and a lack of knowledge on how to prevent the further spread of disease on the individual level are definitely unfavorable when fighting an epidemic.</p><p>The team working on Medy proposed a web platform aggregating verified news related to the epidemic of COVID-19, sharing tips about individual actions that everyone should be performing to stop the spread and stay safe, as well as a system for rewarding users who perform such actions.</p><p>We see the potential for using AI-based solutions in this project. Currently the team assumes that the verification of the news that can be published on the platform would be a manual process. <em>Fake news detection </em>is one of the most popular topics of research in Natural Language Processing. Automation of the verification process, or at least a part of it, would shorten the time between the submission of given information and its publication.</p><h3>EpidemicApp</h3><p>In Poland, patients with symptoms of a disease should report it to a sanitary-epidemiological station by phone. The capacity of such a system is very limited and only a fraction of reports can be processed in a given time, meaning that many patients must wait on hold or redial later without any guarantee that they’ll succeed.</p><p>The team proposed an online submission system for such reports that could later be processed by the personnel of sanitary-epidemiological stations. This idea would certainly reduce the current burden by the simple fact that in a phone-based system a single operator is required for every single patient. In the online solution, this capacity is limited only by the available transfer and processing power of the servers running the system.</p><p>Although the application would massively improve the situation of reporting, the reports still must be processed to decide if a patient is in a high risk group or not. This could be automated by AI. A model could use data from filled forms, even unstructured free form text data, to assess the probability of infection in each case. Documents with the highest probabilities would then be assessed by qualified personnel. Automating the process of initial verification would allow us to identify potentially infected persons much faster and undertake adequate action.</p><p>If you want to read more about AI, check out our site at Tooploox: <a href="https://www.tooploox.com/ai">https://www.tooploox.com/ai</a></p><p><strong>Sources:</strong></p><ol><li><a href="https://www.guaana.com/challenge/dSvc4QnSmsSvmDG5k/results/hGFw22i6o8FmjodPA/8t6bfSKMiSjq3AfWP/main">Detecting Covid-19 on X-rays with deep learning</a></li><li><a href="https://www.guaana.com/challenge/dSvc4QnSmsSvmDG5k/results/hGFw22i6o8FmjodPA/QDmv9PivHPgGbnThk/main">Medy — be rewarded for prevention ✨</a></li><li><a href="https://www.guaana.com/challenge/dSvc4QnSmsSvmDG5k/results/hGFw22i6o8FmjodPA/zpXMejdFaqSBektvZ/main">EpidemicApp</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3572e07a4566" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/top-3-projects-from-the-hack-the-crisis-hackathon-3572e07a4566">Top 3 Projects from the “Hack the Crisis” Hackathon</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You’ve created a video. Will your audience watch it till the end?]]></title>
            <link>https://medium.com/tooploox-ai/youve-created-a-video-will-your-audience-watch-it-till-the-end-66b30d297b26?source=rss-41c110f2afb5------2</link>
            <guid isPermaLink="false">https://medium.com/p/66b30d297b26</guid>
            <category><![CDATA[video-processing]]></category>
            <category><![CDATA[retention]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-video]]></category>
            <dc:creator><![CDATA[Adam Słucki]]></dc:creator>
            <pubDate>Fri, 31 Jan 2020 08:44:43 GMT</pubDate>
            <atom:updated>2020-01-31T08:44:43.101Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_WG5nY6g9_FhrFO1fjIc3A.jpeg" /></figure><p>In today’s world of digital media it’s not enough to get as many views as possible. The attention is definitely important but if the content isn’t engaging, you won’t build a sustainable business on it. But is it possible to predict engagement of your audience before publishing a video and risking a bad first impression? The answer is yes and in this post we’ll describe how to approach this problem with AI.</p><h3>The idea</h3><p>Let’s say you’re responsible for the content creation. You’ve been working on a video and now it’s ready. You’re just not sure if you haven’t included any shots that would prevent your audience from watching the whole piece. Or maybe some shots are too long or the combination of them is off-putting? On the other hand, you’d probably like to know what parts are especially appealing and likely to keep your audience engaged.</p><p>This problem needs a tool that will analyse a video and provide you with relevant information and at the same time allow you to test and edit your content until it’s perfect without trials and errors contaminating your official content feed. It needs an AI solution for predicting drops in retention throughout the whole video.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/999/0*tccvam5byxVLha4M" /><figcaption><strong>Fig 1: </strong>The idea: we have a video that we feed to a neural network. We want the neural network to predict viewers retention from that video.</figcaption></figure><h3>The retention data</h3><p>While you’re watching videos online you’re being watched as well! Or at least some data regarding your interaction with a video player might be recorded and aggregated — where you rewinded a video, what parts you skipped, at what point you stopped watching entirely.</p><p>Based on that data we can assess what parts of a given video are appealing for viewers and what are usually skipped. It’s much more reliable than obtaining the same information with surveys that could be biased by social norms and respondents’ beliefs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/778/1*EIRduIzMEQYc63ch4cgVRQ.png" /><figcaption><strong>Fig. 2: </strong>Example of an interpolated retention curve with a description. Instead of the time in seconds the x axis denotes a time bin.</figcaption></figure><h3>How to compare two very different videos?</h3><p>Now, when you’re familiar with the retention and the problem in general you may wonder if analysing videos in such a way makes any sense. I mean, videos have various lengths and viewers behave differently depending on what part of a video they’re watching. The beginning of a video can determine if you’ll continue watching it or not, but it’s less probable that you’ll stop watching a video if it already sparked your interest. What’s more, it’s easy to skip the last 20 seconds of a 1-minute video but I bet you wouldn’t stop watching a 1-hour movie a few seconds before it ends only because you didn’t like the last scene.</p><p>When comparing videos in our case relativity is the key. What we can conclude here is that the retention drop in this particular video at the time 1:24 is relatively small in comparison with similar drops in all other videos we have in our dataset. What are the similar retention drops? Well, we consider drops from the same value to be similar. It allows us to asign a single number to a given frame in a video — <strong>a percentile of the retention drop</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/959/0*j0vwCp9ZREKjQ1M4" /><figcaption><strong>Fig. 3:</strong> The process: we calculate the retention drop in one video, then we compare it with similar drops in all other videos and check its percentile. <strong>It tells us if our drop is relatively high, which is bad for the video or low, which is desired.</strong></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*l9-3WuL6qWktSwPoBQIAVQ.gif" /><figcaption>This is how the system works. It “scores” fragments of a video according to a predicted drop in retention.</figcaption></figure><h3>What is the input data?</h3><p>We already know what data we want our model to output — percentile of a retention drop. But what should be the basis of its predictions? Definitely a video itself, but let’s be more specific.</p><p>It may seem that the visual layer is the most important in videos as they simply must look attractive. However, in order to be engaging the social media content should convey an important message or tell a story using all the means available to its creators. Content must be easy to consume in every situation. Poor internet connection? No sound? No problem! Many videos have captions or related embedded text. We found that overlayed text is as important for a video success as visual aspects so we decided to include it in our model.</p><p>Videos are also sequential data or to put it simply, they’re sequences of images. So, to predict retention drop in the following second of a video we used a sequence of previous n frames and unique captions overlaying these frames.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/801/0*ADCAc7FPJhX0rfbz" /><figcaption><strong>Fig. 4:</strong> Example of textual overlay on a video frame. We extract such overlays using the method described here: <a href="https://arxiv.org/pdf/1804.10687.pdf">https://arxiv.org/pdf/1804.10687.pdf</a></figcaption></figure><h3>Does the model work at all?</h3><p>How to evaluate the model is a serious question. Of course, there are always metrics. However, it’s really difficult to persuade stakeholders that if the mean squared error equals 10, we can celebrate but if it’s 15, we need more funding. It’s also impossible to look at the model’s output and say “Yep, works perfectly and this fragment definitely was going to cause a huge retention drop”.</p><p>For sure we can’t say that any fragment of a video actually causes any change in retention. But that old mantra saying that correlation doesn’t imply causation inspired us to something else. Maybe we should correlate our predictions with something that the majority of people actually care about! If you’re the one who cares about the mean squared error and can’t guess, I’ll tell you — it’s money.</p><p>The only problem now is that we don’t have any data about earnings. But at least we know that they depend on the number of views. So, the easiest idea is to use retention drop predictions to score each video with a single number (for example, the average drop) and divide videos into two groups using some threshold (median, to be fair). It will give us a set of “good” videos with low retention drops and “poor” videos with high retention drops. Finally, we can calculate the average number of views in each group and compare it (preferably using a statistical test). If the mean in the group of “good” videos shows that they generate more views, it’ll be an indicator that our model is useful and should be deployed on production right away.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/967/0*A0IRGqmIdDA_cauC" /><figcaption><strong>Fig. 5: </strong>Chart showing that “good” videos selected based on retention drop predictions have much more organic views than “poor” videos.</figcaption></figure><h3>So what?</h3><p>We showed that our model may indeed mark correctly fragments of videos that have something to do with their performance in social media. It can also prevent video creators from posting content that the audience won’t like. And on top of that it indirectly correlates with money. Does it mean we found a holy grail of social media? Probably not, but the tool would definitely be a nice add-on to Adobe Premiere.</p><p>If you’re interested in putting AI into your video analysis, whether it’s related to social media or not, feel free to contact us: <a href="https://www.tooploox.com/">https://www.tooploox.com/</a> We’ll be happy to discuss your problem!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66b30d297b26" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/youve-created-a-video-will-your-audience-watch-it-till-the-end-66b30d297b26">You’ve created a video. Will your audience watch it till the end?</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>