<?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 Ankita on Medium]]></title>
        <description><![CDATA[Stories by Ankita on Medium]]></description>
        <link>https://medium.com/@ankita-mit?source=rss-de26d8ccf439------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*rThf9OHYlbrSql3P</url>
            <title>Stories by Ankita on Medium</title>
            <link>https://medium.com/@ankita-mit?source=rss-de26d8ccf439------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 17:22:31 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ankita-mit/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[Creating Art with Neural Networks]]></title>
            <link>https://ankita-mit.medium.com/creating-art-with-neural-networks-9a7ab6053a99?source=rss-de26d8ccf439------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a7ab6053a99</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-art]]></category>
            <dc:creator><![CDATA[Ankita]]></dc:creator>
            <pubDate>Wed, 31 Mar 2021 05:27:58 GMT</pubDate>
            <atom:updated>2021-03-31T17:33:22.676Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/712/1*UxZsmKnDoGtWjbRyiNHnJg.png" /><figcaption>Art created using compositional pattern producing network (CPPN) (Picture Created By: Ankita Singh, Homework for 4.453, MIT)</figcaption></figure><p>Artwork created by machine learning algorithms is an exciting concept. In 2018, Christie’s Prints and Multiples art auction in New York sold a <a href="https://www.christies.com/lot/lot-edmond-de-belamy-from-la-famille-de-6166184/?from=salesummery&amp;intObjectID=6166184&amp;sid=18abf70b-239c-41f7-bf78-99c5a4370bc7">painting</a> created by Generative Adversarial Networks(GAN) for $432,000. The creator of this art piece used a dataset of about 15,000 portraits between the 14th and 20th centuries and trained the Generator segment of their algorithm to create original art that resembled the images it was trained on. The generator segment continuously perfected its work until the other segment of the algorithm, the Discriminator, could not tell apart whether the art was generated by a human or a machine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*7sLRr0jlQuT8pB6X.jpg" /><figcaption>Edmond de Belamy, from <em>La Famille de Belamy, Sold by Christie&#39;s at New York (Image Courtesy: Christie’s)</em></figcaption></figure><p>In this article, I will explain a slightly simpler process of making art with randomly initialized neural networks. These neural networks or also known as Compositional Pattern Producing Networks (CPPNs).</p><h3>How does a CPPN generate art?</h3><p>Mathematically, CPPN can be expressed as a function 𝑓(𝑥,𝑦) that maps a point on a canvas to a color value (1 value for BW, 3 values for RGB). In order to generate an image, a digital canvas is divided into grids and then each point generated by our CPPN function is rendered as a pixel on this digital canvas. Hence, creating an aesthetically pleasing distribution of rendered pixels or an art. This technique is explained in detail in <a href="https://blog.otoro.net/2016/03/25/generating-abstract-patterns-with-tensorflow/">this</a> blog post by David Ha and in this research <a href="http://eplex.cs.ucf.edu/papers/stanley_gpem07.pdf">article</a> by Prof. Kenneth Stanley. The CPPN can be explained graphically in the picture below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/738/0*9vvShKyW4jqkLQVt.png" /><figcaption>CPPN Graphical Representation (Image Courtesy: <a href="http://eplex.cs.ucf.edu/papers/stanley_gpem07.pdf">Compositional Pattern Producing Networks: A Novel Abstraction of Development, Prof. Kenneth Stanley</a>)</figcaption></figure><p>The neural architecture of the CPPN and its weights/parameters define the color and pattern of the image generated. In addition, experiments with CPPN have shown that adding 𝑟² =𝑥²+𝑦² as an input to the network increases the centrality of the generated image, which gives rise to more interesting patterns. A sample neural architecture can be shown in the image below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Wr6Qd8fvMGhkVDCG" /><figcaption>CPPN Neural Architecture Example (Image Courtesy: <a href="https://blog.otoro.net/2016/03/25/generating-abstract-patterns-with-tensorflow/">Generating Abstract Patterns with Tensorflow</a>)</figcaption></figure><p>Below you can see some of the images where I varied the neural architecture of my CPPN to create interesting art patterns.</p><p><strong>Image 1: </strong>The image below generated by Network Architecture 1 is a simpler representation with limited pixel detailing. This indicates that the functions are still simple and relatively easy to explain with only 3 hidden layers and 32 neurons in each layer. I considered this as my base image and then further modified the number of layers, number of neurons, and type of activation function to generate variations of my network architecture.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/0*HoIL4gHgOHysAejs" /></figure><p><strong>Image 2:</strong> The image below generated by Network Architecture 2 started showing some of the function complexity. In this architecture, I increased the hidden layers from 3 to 9 while keeping all others factors the same.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/0*JiNuGC7OrKOJH1aa" /></figure><p><strong>Image 3: </strong>My 3rd network architecture was designed to explore the effects of adding more neurons to the layers. Now comparing Image 3 with Image 2 where I increased the number of neurons per layer from 32 to 128, you will notice that the image is now becoming more smooth. This is because as I added more neurons at each layer, I started seeing more function overlap, and thus the edges are now blending and the pixels are losing their function definition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/0*ExN_gkvXfRtvaEmv" /></figure><h3>Conclusion</h3><p>In conclusion, I was extremely impressed by the power of neural networks to generate extremely complex functions as seen in these images. If you would like to follow and create some of these images, I would recommend following <a href="https://blog.otoro.net/2016/03/25/generating-abstract-patterns-with-tensorflow/">this article</a> by David Ha.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a7ab6053a99" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[UV-Jal — Low-Cost Water Disinfection using UV-C]]></title>
            <link>https://ankita-mit.medium.com/uv-jal-low-cost-water-disinfection-using-uv-c-38b5a6035ba1?source=rss-de26d8ccf439------2</link>
            <guid isPermaLink="false">https://medium.com/p/38b5a6035ba1</guid>
            <category><![CDATA[mit]]></category>
            <category><![CDATA[world-water-day]]></category>
            <category><![CDATA[water-sanitation]]></category>
            <category><![CDATA[uvc-disinfection]]></category>
            <category><![CDATA[water]]></category>
            <dc:creator><![CDATA[Ankita]]></dc:creator>
            <pubDate>Wed, 24 Mar 2021 02:56:38 GMT</pubDate>
            <atom:updated>2021-03-24T02:56:38.786Z</atom:updated>
            <content:encoded><![CDATA[<h3>UV-Jal — Low-Cost Water Disinfection using UV-C</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L3IRreC60O4AV6-oUCc-BA.jpeg" /></figure><p>On March 22, the World celebrated “World Water Day”, a day to appreciate the value of water and raise awareness. This year’s theme for World Water Day was “Valuing Water”. On this occasion, I want to share a project that my friend, Deb, and I have been working on at MIT under the supervision of Prof. Susan Murcott.</p><p>Thousands of families all around the world do not have access to clean water. In India alone, waterborne diseases affect approximately 38 million people every year and over 1.5 million children die due to waterborne diseases like diarrhea. Although a problem of this scale cannot be solved easily, our project focussed on water quality issues in West Singhbhum, a district in the state of Jharkhand in eastern India. Our project focuses on a scalable and affordable compact device to disinfect water using UV-C rays.</p><h3><strong><em>So, How does UV-C water disinfection work?</em></strong></h3><p>The UV-C disinfection technology uses short wave ultraviolet (UV) radiations (100–300 nm) that are known to be fatal to bacteria and viruses. These organisms are inactivated once a sufficiently lethal dose of UV light has been absorbed, altering the molecular structure, preventing reproduction, and thus rendering the microorganisms harmless to humans.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yShiM3FjfVecdm8F" /><figcaption>UV-C Wavelength Range and Its Effect on the Microorganism DNA (Image Courtesy: WEDECO –Xylem Water Solutions Herford GmbH)</figcaption></figure><p>UV dose is measured in microwatt-seconds of UV energy (at or close to 254 nm) per sq. cm of water surface. The dose to inactivate 90% of E. coli is 3000 µWs/cm2. Other pathogenic bacteria and viruses have doses of similar magnitude (rotavirus at 8000 µWs/cm2 is the highest among these).</p><h3>How did we make a UV-C based water disinfection unit?</h3><p>We used mostly off-the-shelf raw materials to create our prototype unit. This included a water-safe UV-C lamp, food-grade PVC pipes and fittings, and a flow control valve. The design ensures UV-C is not exposed to naked eyes which can be harmful to eyes. The most important aspect of the design is its low cost, durability, and simple maintenance. The video below shows a quick and easy run-down of our prototype assembly process.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FTEVJ0oI4B0s%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DTEVJ0oI4B0s&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FTEVJ0oI4B0s%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/c5c505876d0e71c1df87e98e0462d8c5/href">https://medium.com/media/c5c505876d0e71c1df87e98e0462d8c5/href</a></iframe><h3>How did we test it?</h3><p>We began our experimental unit test by collecting water from a source with possible E-Coli contamination. We used 3 different E-Coli/Total Coliform presence tests for our experiments. Two of these tests were a simple +ve/-ve test and one was a quantified measure of E-Coli presence in the water sample.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*poTp6cNGtjXD10_uRYsMsw.png" /><figcaption>Experiments for E-Coli/Total Coliform Presence. ECC-Vial(Top Left), Colliert (Top Middle and Right), and Compact Dry(Bottom)</figcaption></figure><p>In the picture on the top left, the ECC-Vial (developed at MIT) test indicates a light blue color for the presence of E-Coli and a slight pink hue for the presence of total coliforms. In the top middle and right, the colliert test shows the presence of E-Coli (as changed yellow color) and total coliforms (by fluorescence in the presence of UV blacklight). Finally, in the bottom picture we see the compact dry test which is a quantitative measure of E-Coli presence (note that we didn’t find much E-Coli to incubate as visible colonies in this test).</p><h3>Where can it be used?</h3><p>We had a lot of fun building the prototype units in labs and seeing the efficacy of these units in E-Coli disinfection. The simplicity of the design and ease of assembly makes this prototype extremely easy to make in any kind of situation where there is a chance of E-Coli/Total Coliform contamination in water. Most importantly, this can be life-saving in humanitarian missions in places where it&#39;s difficult to get access to a clean water supply.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FlXZ2dKyEfRHvnD_.jpg" /><figcaption>Young girls drink water from a canal flowing between nearby refugee camps and the main road in Peshawar, Pakistan. (Image Courtesy: <em>Majority World/UIG via Getty Images</em>)</figcaption></figure><p>We plan to release this product as an open hardware project, so anyone can manufacture and use the project. We like to thank MIT D-lab, JWAFS, and Prof. Susan Murcott for the resources and support that made this project possible.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=38b5a6035ba1" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>