<?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 raman r on Medium]]></title>
        <description><![CDATA[Stories by raman r on Medium]]></description>
        <link>https://medium.com/@ramanrajarathinam_88127?source=rss-35e0dea2dfdc------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*dKHuSsaaWXa-Y3Vv.</url>
            <title>Stories by raman r on Medium</title>
            <link>https://medium.com/@ramanrajarathinam_88127?source=rss-35e0dea2dfdc------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 04:59:35 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ramanrajarathinam_88127/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[Machine learning: via Bayesian Terms]]></title>
            <link>https://medium.com/@ramanrajarathinam_88127/machine-learning-via-bayesian-terms-9f47b55594c4?source=rss-35e0dea2dfdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f47b55594c4</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[first-post]]></category>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[bayesian-machine-learning]]></category>
            <dc:creator><![CDATA[raman r]]></dc:creator>
            <pubDate>Sat, 15 Sep 2018 14:06:58 GMT</pubDate>
            <atom:updated>2018-09-17T04:41:20.139Z</atom:updated>
            <content:encoded><![CDATA[<p><em>The Crux: </em>With your belief (<em>prior probability</em>) and known evidence/data-points (<em>likelihood</em>), you’re going to update your <em>posterior probability</em>. Initially, start with a belief (Null hypothesis), collect each data point which is going to strengthen or weaken your belief, and you update your <strong>hypothesis</strong> all the time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/865/1*f5s6vgEBMyqoqypnzi77JA.png" /></figure><p>Sounds simple and intuitive? Great.!!</p><p><em>But what is </em><strong><em>Hypothesis?</em></strong></p><p>In machine learning, it can be considered as any set of rules/process, which are given examples<em> </em>or training data, to learn the hidden nature of this process.</p><p>Let’s go through this with a general example,</p><p><strong>Null hypothesis (Hypothesis 0):</strong></p><p>In the 1680s, Newton proposed some set of equations that described some physical properties of the world around us, from Gravity to Motion of the planets. But how does it work?</p><p>Newton was only able to feel the gravity, but he didn’t know how it worked.</p><p><strong>Hypothesis 1:</strong></p><p>In the 19th Century, General Relativity (1915) based on Special Relativity (1905) by Einstein changed the perspective of the fundamental nature. Einstein saw gravity, a Space-Time curvature and space had 3 dimensions. So, technically we have revised/updated our knowledge from the Null Hypothesis to the Hypothesis 1.</p><p><strong>Hypothesis 2:</strong></p><p>In future we may find “Ultimate theory of everything”, that can explain the properties of Quantum Particles to Massive stars. Who knows it may be String theory or Quantum gravity.</p><p>(Wait! Are we in simulation?)</p><p>Back to Machine Learning.</p><p>Okay let’s redefine this process: data is <strong><em>D</em></strong> and hypothesis is <strong><em>h.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/644/1*1Y-uIyIZgu5KCV7w0gXUEg.png" /><figcaption>Baye’s theorem</figcaption></figure><p><strong>P(h/D) — posterior, </strong>Generally, you have many hypotheses to choose from and you want to find ‘which hypothesis is most suitable for your data’.</p><p><strong>P(D/h) — likelihood, </strong>having chosen one hypothesis by belief, you are going to find out ‘how well data fits into this hypothesis’.</p><p><strong>P(h) — prior, </strong>is asking ‘how many times have I chosen this hypothesis for my data?’.</p><p><strong>P(D) — smoothing constant,</strong> you don’t have to care about it, because it doesn’t depend on the hypothesis. 😉</p><p>And all the way, after each iteration, you try to find the best hypothesis.</p><p><em>P.S:</em> <em>If you find</em> any <em>inaccuracies, feel free to point them out, either in the comments or privately.</em></p><p><strong>Where do we really use?</strong></p><p>[1] Neural Networks:</p><p>i) A neural network is a series of layers, with each layer containing nodes and assigned with some weights. You pass your data through these layers and evaluate for correctness. Then the weights of all the neurons involved will be adjusted slightly, reducing the values of the ones that were wrong and increasing the ones that were right.</p><p>ii) Hyperparameter tuning via <strong>Bayesian</strong> <strong>Optimization </strong>(Mixture Gaussian Process and Bayesian Inference).</p><p>[2] <a href="https://en.wikipedia.org/wiki/Latent_Dirichlet_allocation">Latent Dirichlet Allocation</a>: A popular topic modeling approach has two variants, Gibbs Sampling and <strong>Bayesian Inference</strong>.</p><p>[3] In Information retrieval, we use <strong>Bayesian Network </strong>(probabilistic graphical network) to find a relation between terms and documents.</p><p>Reference:</p><p>[1] <a href="https://www.youtube.com/watch?v=R13BD8qKeTg&amp;t=2s">https://www.youtube.com/watch?v=R13BD8qKeTg&amp;t=2s</a></p><p>[2] <a href="https://towardsdatascience.com/when-bayes-ockham-and-shannon-come-together-to-define-machine-learning-96422729a1ad">https://towardsdatascience.com/when-bayes-ockham-and-shannon-come-together-to-define-machine-learning-96422729a1ad</a></p><p>[3] <a href="http://fastml.com/bayesian-machine-learning/">http://fastml.com/bayesian-machine-learning/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f47b55594c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>