<?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 Akanksha Paul on Medium]]></title>
        <description><![CDATA[Stories by Akanksha Paul on Medium]]></description>
        <link>https://medium.com/@akanksha-paul?source=rss-4bf50a1b4da------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*3-XUh_wZidCOdIilJlub2w.jpeg</url>
            <title>Stories by Akanksha Paul on Medium</title>
            <link>https://medium.com/@akanksha-paul?source=rss-4bf50a1b4da------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 22:51:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@akanksha-paul/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[Interview Experience for RSDE at Amazon]]></title>
            <link>https://akanksha-paul.medium.com/interview-experience-for-rsde-at-amazon-9bdf3568428?source=rss-4bf50a1b4da------2</link>
            <guid isPermaLink="false">https://medium.com/p/9bdf3568428</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[interview-questions]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[applied-scientist]]></category>
            <category><![CDATA[amazon]]></category>
            <dc:creator><![CDATA[Akanksha Paul]]></dc:creator>
            <pubDate>Sat, 28 Nov 2020 17:40:34 GMT</pubDate>
            <atom:updated>2020-11-28T17:40:34.955Z</atom:updated>
            <content:encoded><![CDATA[<p>The interview process comprises of two initial phone screen rounds, followed by face-to-face interactions with a bunch of Applied Scientists, Principal Applied Scientists, and SDE’s.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/259/1*RLv9tvexWfJ_eBUkpHV-7g.jpeg" /></figure><p>The process is rigorous and requires:</p><ul><li>Basic knowledge of Data Structures and Algorithms</li><li>Machine Learning Breadth</li><li>Machine Learning Depth</li><li>Machine Learning applications</li><li>Behavioral interview to see cultural fit</li></ul><p><strong>INITIAL PHONE SCREEN ROUNDS</strong></p><p>These rounds examine the breadth and depth of the Machine Learning Basics. The interviews would kick-off with a basic introduction of the interviewee and the interviewer. A follow-up would be 20–25 minute discussion about the projects mentioned in the Resume. Expect questions about the importance of the project and its applications. Further, there would be a question about the Machine Learning paradigms that you have utilized. Tell only the models, that you have in-depth knowledge about to answer the questions. I had worked around Decision trees, CNN’s, LSTM’s, VAE’s, and GAN’s. Below are the questions asked to me over the two screening rounds:</p><ul><li>Merge two sorted lists.</li><li>How to merge K-sorted lists? Use heap for optimal time complexity.</li><li>Find the largest contiguous sum of an array containing positive and negative numbers (Kadane’s algorithm).</li><li>What is the difference between bagging and boosting?</li><li>What do high bias and high variance mean? Why does it occur?</li><li>What are random forests? Why are they random?</li><li>Why do we need LSTM’s? Why can’t simple neural networks work?</li><li>Why we require attention in LSTM?</li><li>How do you represent the text?</li><li>Why does dropout help in regularization? How does it work for Neural Networks? How is it implemented for LSTM’s?</li><li>A user would navigate through the Amazon website to search for the required product. Given the search text of the user, how would you recommend the products to the user? The dataset creation itself is a big task here, as there would be millions of queries. How would you create the dataset?</li><li>What is a triplet loss? How should the negative samples be chosen?</li></ul><h3>Onsite rounds</h3><p>Once the initial phone screening rounds are cleared, the recruiter will arrange for the onsite interviews. These rounds examine not only the breadth and depth of Machine Learning. But also look at a broader aspect of applied machine learning in the business scenario. The basic knowledge of data structures and algorithms is naturally expected. Below is the list of questions asked to me over the different rounds:</p><p><strong><em>Round 1: Algorithms and Data Structures</em></strong></p><ul><li>Remove all spaces from the string. Example:<br>Input: “this is the coding round”<br>Output: “thisisthecodinground”</li><li>Remove all spaces from the string in a mannered way. Example:<br>Input: “this is the coding round”<br>Output: “this is the coding round”</li><li>There are a total of n tasks, labeled from 0 to n-1. Some tasks may have prerequisites tasks, for example, to pick task 0 you have to first finish tasks 1, which is expressed as a pair: [0, 1].<br>Given the total number of tasks and a list of prerequisite pairs, return the ordering of tasks to finish all tasks.<br>Input: [[1, 0], [2, 0], [3, 1], [3, 2]]<br>Output: [0, 1, 2, 3] or [0, 2, 1, 3]</li></ul><p><strong><em>Round 2: Machine Learning Breadth and Depth</em></strong></p><ul><li>Akanksha noticed that during every given minute, there is a 1/500 chance that her Medium post gets a ‘like’, independently of what happens during another minute. How do you model this distribution and why?</li><li>Follow-up, if S be the total number of Likes that Akanksha receives during 24 hour period. What is the probability of S=0?</li><li>What is the difference between Bagging and Boosting?</li><li>What is Dropout regularization? Is it more similar to L1-regularization or L2-regularization?</li><li>When to prefer L2-regularization?</li><li>What is multicollinearity? How do you detect for multi-collinearity?</li><li>What is Bias Variance tradeoff for a decision tree?</li><li>What are the ways to reduce variance wrt to the decision trees? (Random forests)</li><li>How does random forest reduce variance without increasing bias?</li><li>What is KL divergence? What are the other divergence measures?</li><li>What are the different performance measures utilized in various projects?</li><li>Why there is a need for a convolution layer in CNN’s? What if it was a fully connected layer?</li><li>Why do we need strides?</li><li>What is the idea of 1*1 convolutions?</li><li>Describe GAN’s.</li></ul><p><strong><em>Round 3: Machine Learning Breadth and Depth</em></strong></p><ul><li>Why does dropout help in regularization? How do you implement it?</li><li>What is Linear Regression? What is the loss function for optimization?</li><li>What are the different ways to optimize loss function? When to prefer gradient Descent over Analytical Solution?</li><li>Bias Variance tradeoff wrt Linear Regression. How does the graph vary concerning the complexity of the model?</li><li>What are the ways to regularize the Linear regression model?</li><li>Which is easier to optimize? L1-regularization or L2-regularization?</li><li>When to prefer L1- regularization or L2- regularization?</li><li>What are other ways of regularization?</li><li>Why does dropout help in regularization? How do you implement it?</li><li>Now consider implementing CNN’s, what extra regularization methods you can think?</li><li>If a small dataset of images is given, how do you train CNN’s? How is the model designed, what are the various training parameters?</li><li>If the small dataset above, is imbalanced? How do you improve the biasness of the model?</li><li>What is GAN’s? What is the loss function corresponding to it? What are the problems of GAN’s?</li><li>Why do we require an autoencoder?</li><li>How does VAE improve over AE? What is the loss function utilized in VAE?</li><li>What is KL divergence?</li><li>What to prefer Forward KL or Reverse KL?</li></ul><p><strong><em>Round 4: Behavioral Round</em></strong></p><p>Amazon examines every interviewee on the cultural fit on the basis of their 14 leadership principles. These principles should be demonstrated via past projects or experiences in your life.</p><ul><li>Describe the project. What do you think is the improvement of the project?</li><li>Why are you looking for a change?</li><li>What is the newest cool thing you have learned?</li><li>What do you do if you are stuck at a problem? How do you resolve it?</li><li>Implement two functions:</li></ul><ol><li>songPlayed(artistName,songName): that takes as input aristName and songName, and updates the number of times a song is played.</li><li>getTopSong(artistName): that takes the artistName and gets the most frequently played song of the artist.</li></ol><p>We can uniquely identify an artist’s name. But a song name can be identified uniquely with an artist’s name, i.e. no unique song name. Both the functions should take O(1) time.</p><p>In the end, just be confident when you walk into the interview room and give it your best shot.</p><h3>Thanks for reading…. Hope you find it beneficial….</h3><p>Do upvote, if you find this helpful!! Thanks. 😊</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9bdf3568428" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Interview Experience for Applied Scientist at Amazon]]></title>
            <link>https://akanksha-paul.medium.com/interview-experience-for-applied-scientist-at-amazon-5ff5003375dd?source=rss-4bf50a1b4da------2</link>
            <guid isPermaLink="false">https://medium.com/p/5ff5003375dd</guid>
            <category><![CDATA[amazon]]></category>
            <category><![CDATA[applied-scientist]]></category>
            <category><![CDATA[interview-questions]]></category>
            <category><![CDATA[interview]]></category>
            <dc:creator><![CDATA[Akanksha Paul]]></dc:creator>
            <pubDate>Fri, 12 Jun 2020 19:01:57 GMT</pubDate>
            <atom:updated>2020-06-12T19:01:57.862Z</atom:updated>
            <content:encoded><![CDATA[<p>The interview process comprises of two initial phone screen rounds, followed by face-to-face interactions with a bunch of Applied Scientists, Principal Applied Scientists, and Manager.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/321/1*3jqMeCMJ-fz8UtzJFP_sNQ.png" /></figure><p>The process is rigorous and requires basic knowledge of:</p><ul><li>Data Structures and Algorithms</li><li>Machine Learning Breadth</li><li>Machine Learning Depth</li><li>Machine Learning Applications</li><li>Scalable Machine Learning</li></ul><h4>Initial phone screen rounds</h4><p>These rounds examine the breadth and depth of the Machine Learning Basics. The interviews would kick-off with a basic introduction of the interviewee and the interviewer. A follow-up would be 20–25 minute discussion about the projects mentioned in the Resume. Expect questions about the importance of the project and its applications. Further, there would be a question about the Machine Learning paradigms that you have utilized. Tell only the models, that you have in-depth knowledge about to answer the questions. I had worked around Decision trees, CNN’s, LSTM’s, VAE’s, and GAN’s. Below are the questions asked to me over the two screening rounds:</p><ul><li>How is a decision tree created? What is mutual information?</li><li>Explain why you need VAE’s over Autoencoders. What is the loss you minimize?</li><li>What is batch normalization? How does a batch normalization layer help?</li><li>Difference between LSTM’s and GRU’s.</li><li>What does peephole keyword in TensorFlow mean for LSTM’s?</li><li>What are the different types of text representations used? What is the difference between Word2Vec, Glove, FastText?</li><li>What is the difference between Eigen Value Decomposition (EVD) and Singular Value Decomposition (SVD)? When does SVD behave the same as EVD?</li><li>What is the difference between bagging and boosting?</li><li>What is GBDT? (Gradient Boosted Decision Trees)</li><li>Why there is a need for padding in CNN’s?</li><li>Explain ResNet architecture.</li><li>Difference between Logistic Regression and Linear Regression.</li><li>What does maximizing the likelihood function for Logistic Regression mean? What happens if we include some prior over the weights of Logistic Regression? Why we assume prior over the weights?</li><li>When does MLE equal to MAP estimate?</li><li>Why is it preferred to perform standardization over the features?</li></ul><h4>Onsite rounds</h4><p>Once the initial phone screening rounds are cleared, the recruiter will arrange for the onsite interviews. These rounds examine not only the breadth and depth of Machine Learning. But also look at a broader aspect of applied machine learning to scalable systems in the business scenario. The basic knowledge of data structures and algorithms is naturally expected. Below is the list of questions asked to me over the different rounds:</p><p><strong><em>Round 1: Scalable ML and Applications</em></strong></p><ul><li>When the users are navigating through the Amazon website, they are viewing a product, clicking a product, and then purchasing a product. Given a user and a product, model a system that calculates the probability of viewing, clicking, and purchasing the product.</li><li>How the dataset can be created? What features are required for the user and the product? Would you prefer three models that model each view, clicking, and purchasing probability independently or jointly?</li><li>The Amazon dataset has millions of data points. How would you go about training such a model?</li><li>What are the different ways of down sampling the data?</li><li>What should be the properties of the down sampled data?</li><li>If you want to sample 10 points out of each batch of 1000 points. Toss a coin and if the probability is less than 0.01 then keep the point, else discard it. What is the distribution of this model? What is its mean and variance?</li><li>How is this sampling different from random sampling?</li><li>What is reservoir sampling?</li><li>Create a monthly Amazon cart for the user. How would you model this problem?</li></ul><p><strong><em>Round 2: Machine Learning Breadth and Depth</em></strong></p><ul><li>What is the difference between modeling the problem as P(y|x) and P(y,x)? What are the advantages of one over the other? Which algorithms are usually used and why?</li><li>What is the difference between MLE and MAP wrt to Linear Regression? What does linear regression assume for P(y|x)? What does Logistic regression assume for P(y|x)?</li><li>What is the Bayes Rule? Explain the terms posterior, likelihood, and prior.</li><li>What is marginalization? Why is it required to marginalize over a variable?</li><li>Why is it difficult to calculate posterior in VAE’s? What are the different approaches to calculate the posterior distribution?</li><li>What is batch gradient descent, minibatch gradient descent, and stochastic gradient descent? Which one would you prefer over the other?</li><li>Why does stochastic gradient descent oscillate towards the local minima?</li><li>How the above optimization methods can be improved? Why does momentum term help?</li><li>Do you know anything about Hessian? Can they be utilized for faster training? What are its disadvantages?</li><li>What is the idea of an adaptive learning rate? Describe any adaptive learning methods.</li><li>Describe an instance where you went out of your way to help someone.</li></ul><p><strong><em>Round 3: Data structures and Algorithms</em></strong></p><ul><li>Find all pairs of indices that sum to a target sum. (for both sorted and unsorted array)</li><li>Find the longest substring that is repeated in the string. The substring can be overlapping. Example:<br>Input: banana<br>Output: ana</li></ul><p><strong><em>Round 4: Behavioral Round</em></strong></p><p>A behavioral interview is conducted to access cultural fit of the person with Amazon. Here, a person is expected to know the 14 Leadership Principles of Amazon and should demonstrate some of them in the interview process. Some of the questions that I remember were:</p><ul><li>Describe a time where you did something more than your expected responsibilities.</li><li>What do you think is important short term goals or long term goals? Why?</li><li>Give an instance of a long term goal where you had to sacrifice a short-term goal.</li><li>What does your work mean to you?</li><li>Why did you choose to work in Machine learning?</li></ul><p>In the end, just be confident when you walk into the interview room and give it your best shot.</p><h3><strong>Thanks for reading…. Hope you find it beneficial….</strong></h3><p>If you wish to find some answers to the questions, feel free to post it in comments. Always happy to help… :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5ff5003375dd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>