<?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 Deepika Rana on Medium]]></title>
        <description><![CDATA[Stories by Deepika Rana on Medium]]></description>
        <link>https://medium.com/@deepika_rana?source=rss-65025b331d02------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*PgWFp2T1NlkPzD0CEG_v6w.png</url>
            <title>Stories by Deepika Rana on Medium</title>
            <link>https://medium.com/@deepika_rana?source=rss-65025b331d02------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 18:18:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@deepika_rana/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[Developer Student Club MAIT]]></title>
            <link>https://medium.com/dscmait/developer-student-club-mait-8cfe2a2fec0d?source=rss-65025b331d02------2</link>
            <guid isPermaLink="false">https://medium.com/p/8cfe2a2fec0d</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[community]]></category>
            <category><![CDATA[developer]]></category>
            <category><![CDATA[google-developer]]></category>
            <category><![CDATA[cd]]></category>
            <dc:creator><![CDATA[Deepika Rana]]></dc:creator>
            <pubDate>Sun, 27 Sep 2020 05:56:54 GMT</pubDate>
            <atom:updated>2020-09-27T06:00:45.547Z</atom:updated>
            <content:encoded><![CDATA[<h4>2020–21</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FNSxS70j4GXrgCF5KJz1fw.png" /></figure><p><a href="https://developers.google.com/community/dsc">Developer Student Clubs</a> are university-based community groups for students interested in Google technologies and the latest developer technologies. Students from all undergraduate or graduate programs with an interest in growing as a developer are welcome.</p><h4>Vision</h4><p>To bridge the gap between theory and practical applications for student developers.</p><ol><li><strong>Connect</strong><br>Meet other students on campus interested in developer technologies. All are welcome, including those with diverse backgrounds and different majors.</li><li><strong>Learn</strong><br>Learn about a wide range of technical topics where new skills are gained through hands-on workshops, in-person training, and project building activities.</li><li><strong>Grow</strong><br>Apply your new learnings and connections to build great solutions for local problems. Advance your skills, career, and network. Give back to your community by helping others learn as well.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1009/1*LM7CcoSfmYxfWI-7QD7Wyw.png" /><figcaption>Activites by DSC MAIT</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1010/1*g4lPlcZed7RHP8s8iFokgg.png" /></figure><h3>DSC MAIT Sessions</h3><ul><li>Expert Tech -talks</li><li>Solution challenge</li><li>Hands-on workshop</li><li>Explore ML</li><li>Info sessions</li><li>Speaker sessions</li><li>Bootcamps</li></ul><h4>Domains</h4><ol><li>Web Development</li><li>App Development</li><li>UI/UX Design</li><li>Voice User Interfaces</li><li>Machine Learning</li></ol><h4>How the DSC MAIT community members benefit?</h4><p>Members can learn the latest technologies, make projects in teams or individually, participate in contests, find like-minded people, receive guidance, contribute to open source, and most importantly you’ll <strong>stay aware of opportunities</strong> like international competitions, internships, scholarships, exclusive webinars/workshops…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/285/0*lZ69h-SfD4qVOBxl" /></figure><p>Excited?</p><p><strong>To join our community become a part of our </strong><a href="https://dsc.community.dev/maharaja-agrasen-institute-of-technology-delhi/"><strong>DSC MAIT chapter</strong></a><strong>.</strong></p><p>Follow us for further updates</p><p><a href="https://www.linkedin.com/company/developer-students-club-mait/?viewAsMember=true">LinkedIn</a></p><p><a href="https://twitter.com/dsc_mait">Twitter</a></p><p><a href="https://www.instagram.com/dsc_mait/?igshid=1o7wm0woachko">Instagram</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8cfe2a2fec0d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dscmait/developer-student-club-mait-8cfe2a2fec0d">Developer Student Club MAIT</a> was originally published in <a href="https://medium.com/dscmait">DSC MAIT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Bias and Variance]]></title>
            <link>https://medium.com/udacity-bertelsmann-ai-scholars/understanding-bias-and-variance-81c55448a2e2?source=rss-65025b331d02------2</link>
            <guid isPermaLink="false">https://medium.com/p/81c55448a2e2</guid>
            <category><![CDATA[variance]]></category>
            <category><![CDATA[overfitting]]></category>
            <category><![CDATA[bias]]></category>
            <category><![CDATA[underfitting]]></category>
            <dc:creator><![CDATA[Deepika Rana]]></dc:creator>
            <pubDate>Tue, 11 Feb 2020 15:32:59 GMT</pubDate>
            <atom:updated>2020-06-11T15:45:23.616Z</atom:updated>
            <content:encoded><![CDATA[<p>Applied deep learning is a <em>highly iterative process</em> where a cycle is repeated many times to find them a good choice of network for your application. One of the things that determine how quickly progress is made is the efficiency around the cycle. And splitting the data in terms of train, development and test sets can increase the efficiency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0mX9VZDCmb4bwyI4KWiSVg.png" /></figure><p>Suppose we have a data set as shown in the figure. If a straight line is fit to the data, maybe logistic regression fit to that. This is not a very good fit to the data. And so this is class of <strong>high bias</strong>, so this is <strong>underfitting</strong> the data as shown <em>figure(a)</em>. On the opposite end, if an incredibly complex classifier, maybe deep neural network, or neural network is made to fit, with all the hidden units, maybe the data is fit perfectly, but that doesn’t look like a great fit either. So there’s a classifier of <strong>high variance</strong> and this is <strong>overfitting</strong> the data as shown in <em>figure(c)</em>. And there might be some classifier in between, with a medium level of complexity, that maybe fits it correctly like that. That looks like a much more <strong>reasonable fit</strong> to the data, so that is just right. It’s somewhere in between as shown in <em>figure(b)</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L55wamh_15Ortu8jrqm_XQ.png" /></figure><p>Considering an example of dog picture classification, where a dog is a positive example and a cat is a negative example, the two key numbers to look at to understand bias and variance will be the train set error and the dev set or the development set error.</p><p><em>For the sake of argument, assuming that recognizing dogs in pictures, is something that the human eye can do nearly perfectly, right?</em></p><p>Hypothetically let’s say, the <strong>training set error</strong> is <strong><em>1%</em></strong> and the <strong>dev set error</strong> is <strong><em>11%</em></strong><em>.</em> So in this example, the algorithm performs well on the training set but doing relatively poorly on the development set. This is overfitting the training set, as it does not generalize well, to this whole cross-validation set in the development set. This implies that the assumed example has a <strong>high variance</strong>. Hence, by looking at the training set error and the development set error, one would be able to render a diagnosis that the algorithm has high variance.</p><p>Now, let us say for a different algorithm, <strong>training set error</strong> is <strong><em>14%</em></strong> and <strong>dev set error</strong> is <strong><em>15%</em></strong>. In this case, assuming that humans achieve roughly 0% error, then humans can look at these pictures and just tell if it’s a dog or not, then it looks like the algorithm is not even doing very well on the training set. This is not even fitting the training data that well, then this is underfitting the data. This implies that the algorithm has a <strong>high bias</strong>. But in contrast, this is actually generalizing at a reasonable level to the dev set, whereas performance in the dev set is only 1% worse than performance in the training set. But still, the problem of high bias persists.</p><p>Assuming another example, there is a <strong><em>14%</em></strong> <strong>training set error</strong>, so that’s pretty high bias, but evaluating the <strong>dev set error </strong>it does even worse, around <strong><em>30%</em></strong>. In this case, this algorithm is having a <strong>high bias</strong>, because it’s not doing that well on the training set, and <strong>high variance</strong>. So this has really the worst of both worlds.</p><p>And, if there is a <strong><em>0.5%</em></strong> <strong>training set error</strong>, and <strong><em>1%</em></strong> <strong>dev set error</strong>, then the dog classifier with only 1% error. This implies that the algorithm has <strong>low bias</strong> and <strong>low variance</strong>.</p><blockquote>This analysis is predicated on the assumption, that human level performance gets nearly 0% error or, more generally, that the optimal error, sometimes called base error, so the base in optimal error is nearly 0%.</blockquote><p>Looking at training set error and dev set error helps in diagnosing whether the algorithm has a bias or a variance problem, or maybe both. It turns out that this information helps in much more systematically going about improving the algorithms’ performance.</p><p>When training a neural network the path to be followed is :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1022/1*TGzGFE76OCheYKG3T8EuYA.png" /></figure><blockquote>After having trained an initial model, first thing to consider is does the algorithm have high bias?</blockquote><p>If it does have a high bias, it does not even fit in the training set that well, some things to try is to pick a network, such as more hidden layers or more hidden units, or training it for a longer time. This may work, maybe it won’t. But there are a lot of different neural network architectures and maybe a <strong><em>new network architecture</em></strong> can be used that’s better suited for this problem. Maybe it works, or maybe not. Whereas getting a <strong><em>bigger network</em></strong> almost always helps, <strong><em>training longer</em></strong> doesn’t always help. So when training a learning algorithm, these things can be tried at least to get rid of the bias problems and go back after this and it is done until it fits, at least, fit the training set pretty well. And usually, if there is a big enough network, it usually fits the training data. But if the image is very blurry, it may be impossible to fit it. But if at least a human can do well on the task, if the base error is not too high, then by training a big enough network, then the algorithm performs well at least on the training set. To at least fit or overfit the training set.</p><blockquote>Once the bias is reduced to acceptable amounts then consider, does the have a variance problem?</blockquote><p>And so to evaluate development set performance is to be evaluated. If there is high variance, well, the best way to solve a high variance problem is to get more data. But sometimes <strong><em>more data</em></strong> is not available. Or <strong><em>regularization</em></strong> can be tried to reduce overfitting. But finding a more appropriate <strong><em>neural network architecture</em></strong>, sometimes that can reduce the variance as well as reduce the bias can help. It’s harder to be systematic. So these things need to be tried and kind of keep going back, until there is an algorithm with both low bias and low variance, whereupon it is done.</p><p>This gives the basic structure of how to organize a machine learning problem to diagnose bias and variance, and then to <em>select the right operation</em> to make progress on the problem.</p><p><strong><em>References:</em></strong></p><p><em>Coursera</em> (deeplearning.ai)</p><p>Udacity</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=81c55448a2e2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/udacity-bertelsmann-ai-scholars/understanding-bias-and-variance-81c55448a2e2">Understanding Bias and Variance</a> was originally published in <a href="https://medium.com/udacity-bertelsmann-ai-scholars">Udacity-Bertelsmann AI Track Scholars</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ABCs of Recurrent Neural Networks]]></title>
            <link>https://medium.com/udacity-bertelsmann-ai-scholars/abcs-of-recurrent-neural-networks-101482d91f66?source=rss-65025b331d02------2</link>
            <guid isPermaLink="false">https://medium.com/p/101482d91f66</guid>
            <category><![CDATA[lstm]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[rnn]]></category>
            <dc:creator><![CDATA[Deepika Rana]]></dc:creator>
            <pubDate>Wed, 05 Feb 2020 11:54:44 GMT</pubDate>
            <atom:updated>2020-06-11T15:45:38.221Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pIY1eGo2ONLwCcCZaR1Q6w.png" /></figure><p><strong><em>RNNs </em></strong><em>are designed in such a way that it learns from sequences of data by passing the hidden state from one step in the sequence to the next step in the sequence, combined with the input. </em><strong><em>LSTMs</em></strong><em> are an improvement for RNNs, and are used when the neural network needs to switch between remembering recent things, and things from long time ago.</em></p><p>Suppose there is a neural network which recognizes images and the following image is fitted. And the neural network guesses that the image is most likely a dog with <em>80%</em> probability, a wolf with <em>15%</em> probability and a goldfish with <em>5% </em>probability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_kyjhVgpxEfZUuXRyC88EA.png" /></figure><blockquote><em>But what if the image is actually a wolf ? How would a neural network know?</em></blockquote><p>Let us assume there is a show and the previous image before the wolf was a bear and the previous one was a fox. So, in this case, this information is a hint that the last image is a wolf and not a dog. Each image is analysed with the same copy of a neural network. But output of the neural network is used as a part of the input of the next one which <em>improves the accuracy</em>.</p><p><em>Mathematically speaking,</em> the vectors are combined in a linear function which will then be squished with an activation function. In this way the previous information is used and the final neural net will know that the show is about wild animals and this information is used to correctly predict that the image is of a wolf and not a dog. And this is how a neural network works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HBR0h-pbGTnS7r7qnwhyYw.png" /></figure><p><em>But wait, it has a shortcoming.</em></p><p>Assuming the bear appeared a while ago and the two recent images are a sheep and a rabbit. Based on those two the network does not know if the new image is a dog or a wolf. Since sheep and rabbit are just associated to both domestic and wild animals. So, the information about being in the forest comes all the way back from the bear.</p><p><em>But information getting in is repeatedly squished by sigmoid functions and even worse training a network using backpropagation all the way back, will cause problems such as </em><strong><em>vanishing gradient</em></strong>. At this point most of the information about the bear has been lost as the memory that is stored is normally short term memory. This is where <strong>LSTMs</strong> (Long Short Term Memory) plays an important role.</p><p>Summarizing, how an RNN works:</p><blockquote>Memory comes in and combines with a current event.</blockquote><blockquote>Output comes out as a prediction of what is the input</blockquote><blockquote>Output also comes as a part of the input for the next iteration of the neural network.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iSyHCvh4scqs1WTcFHpuRg.png" /></figure><p><strong>References :</strong></p><p><em>Udacity</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=101482d91f66" width="1" height="1" alt=""><hr><p><a href="https://medium.com/udacity-bertelsmann-ai-scholars/abcs-of-recurrent-neural-networks-101482d91f66">ABCs of Recurrent Neural Networks</a> was originally published in <a href="https://medium.com/udacity-bertelsmann-ai-scholars">Udacity-Bertelsmann AI Track Scholars</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>