 Vimarsh Karbhari

– Engineering Manager

– Medium Acing AI的作者

Google Google拥有全世界顶尖的AI研究科学家，数据工程师和数据科学家。Google的CEO, Sundar Pichai希望能引领着Google成为一流的AI公司，并且已经将AI技术融入到其所有或大部分产品中，从Gmail到拥有大量数据的自动驾驶系统。

Google 的技术面试流程是标准的技术面试流程，由电话视频面试和现场面试组成。

Google AI的阅读准备材料

1. TensorFlow: A system for Large Scale Machine learning.

2. Tools that Google uses both Hardware and Software: AI Tools

3. Unofficial Google Data Science Blog

• What is the derivative of 1/x?
• Draw the curve log(x+10)
• How to design a customer satisfaction survey?
• Tossing a coin ten times resulted in 8 heads and 2 tails.
• How would you analyze whether a coin is fair? What is the p-value?
• You have 10 coins. You toss each coin 10 times (100 tosses in total) and observe results. Would you modify your approach to the the way you test the fairness of coins?
• Explain a probability distribution that is not normal and how to apply that?
• Why use feature selection? If two predictors are highly correlated, what is the effect on the coefficients in the logistic regression? What are the confidence intervals of the coefficients?
• K- mean and Gaussian mixture model: what is the difference between K-means and EM?
• When using Gaussian mixture model, how do you know it is applicable? (Normal distribution)
• If the labels are known in the clustering project, how to how to evaluate the performance of the model?
• You have a google app and you make a change. How do you test if a metric has increased or not?
• Describe the process of data analysis?
• Why not logistic regression, why GBM?
• Derive the equations for GMM.
• How would you measure how much users liked videos?
• Simulate a bivariate normal
• Derive variance of a distribution
• How many people apply to Google per year?
• How do you build estimators for medians?
• If each of the two coefficient estimates in a regression model is statistically significant, do you expect the test of both together is still significant?

Apple AI 被包含在苹果硬件之上的软件中。也就是说，人工智能是苹果的一种服务。根据他们在 2018 年一季度的收益报告，他们的服务收入比去年增长了 18%。截至去年 12 月底，所有服务产品的付费用户数量都超过了 2.4 亿。

Apple的AI和Data Scientist的面试题有很多Hadoop相关的问题，似乎Apple的数据挖掘后端是基于Hadoop，这和很多其他公司不一样。

Apple AI的阅读准备材料

1. How Apple personalized Siri invocation: Personalized Hey Siri

2. ML Journal(Blog by Apple Engineers for ML): Machine Learning Journal

3. Github Libraries(For Development of Custom ML Models): Turi Create

• How do you take millions of users with 100’s of transactions each, amongst 10k’s of products and group the users together in meaningful segments?
• We do pre-screening on the data to remove fraud threats — so how do we find a data sample that we can use to determine a real representation of fraud events?
• Given a table with 1B of user ID and product IDs that the users bought, and another table with product ID mapped with product name. We are trying to find the paired products that are often purchased together by the same user, such as wine and bottle opener, chips and beer. How to find the top 100 of these co-existed pairs of products?
• Describe for me in detail the difference between L1 and L2 regularization, specifically as regards the difference in their impact on the model training process itself.
• Suppose you have 100,000 files spread across multiple servers and you wanted to process all of them? How would you do that in Hadoop?
• What is the difference between Python and Scala?
• Explain LRU Cache.
• How would you design a client-server model where the client sends location data every minute?
• How would you transfer data from one Hadoop cluster to another?
• What are different types of memories in Java?
• How can you handle the daily tedious tasks that go hand in hand with processing metadata for hundreds of titles?
• In terms of data flow and accessibility, how do you measure success in a hidden time frame where the nucleus overloads the border structure of the over complicated file system that redirects computer energy to the cellar dome?
• If you could have one superpower, what would it be?
• You have time series of sensors, predict the next reading.
• Create market basket output using SQL.
• What is your experience with psychophysical experiments?(Research Portfolio based question)
• What is your expertise in characterization? What do you usually use that for? How did you use that in your research and find interesting results?(Research Portfolio based question)
• How do you deal with failure analysis?
• Check if a binary tree is a mirror image on left and right sub-trees.
• What is a random forest? Why is Naive Bayes better?
Microsoft Microsoft AI的阅读准备材料

1. Microsoft AI School: Different Learning Paths

2. AI Demos(Showcases the Data Presentation and Visualization): AI Demos

3. Microsoft Azure AI Solutions (Similar to Amazon AWS): Projects

4. Microsoft Research Podcast: Research Podcast (Courtesy: Petercooper via HackerNews)

• Merge k (in this case k=2) arrays and sort them.
• How best to select a representative sample of search queries from 5 million?
• Three friends in Seattle told you it’s rainy. Each has a probability of 1/3 of lying. What’s the probability of Seattle is rainy?
• Can you explain the fundamentals of Naive Bayes? How do you set the threshold?
• Can you explain what MapReduce is and how it works?
• Can you explain SVM?
• How do you detect if a new observation is outlier? What is a bias-variance trade off ?
• Discuss how to randomly select a sample from a product user population.
• How do you implement autocomplete?
• Describe the working of gradient boost.
• Find the maximum of sub sequence in an integer list.
• What would you do to summarize a twitter feed?
• Explain the steps for data wrangling and cleaning before applying machine learning algorithms.
• How to deal with unbalanced binary classification?
• How to measure distance between data point?
• Define variance.
• What is the difference between box plot and histogram?
• How do you solve the L2-regularized regression problem?
• How to compute an inverse matrix faster by playing around with some computational tricks?
• How to perform a series of calculations without a calculator. Explain the logic behind the steps.
• What is a difference between good and bad Data Visualization?
• How do you find percentile? Write the code for it.
• Find max sum subsequence from a sequence of values.
• What are the different regularization metrics L1 and L2?
• Create a function that checks if a word is a palindrome.
Amazon Amazon利用 Alexa 部署深度语言学习功能，并通过 AWS 为 AI 提供云基础架构。它还在 Amazon.com 上大规模地构建和部署了世界上第一批推荐系统。

1. AWS Sagemaker(Video): Build, Train and Deploy ML models at scale.

2. Deep Learning AMIs: Tutorials to use AMIs on AWS.

3. Amazon AWS Blog: ML Blog (Great examples on different solutions to data science stack related problems)

• How does a logistic regression model know what the coefficients are?
• Difference between convex and non-convex cost function; what does it mean when a cost function is non-convex?
• Is random weight assignment better than assigning same weights to the units in the hidden layer?
• Given a bar plot and imagine you are pouring water from the top, how to qualify how much water can be kept in the bar chart?
• What is Overfitting?
• How would the change of prime membership fee would affect the market?
• Why is gradient checking important?
• Describe Tree, SVM, Random forest and boosting. Talk about their advantage and disadvantages.
• How do you weight 9 marbles three times on a balance scale to select the heaviest one?
• Find the cumulative sum of top 10 most profitable products of the last 6 month for customers in Seattle.
• Describe the criterion for a particular model selection. Why is dimension reduction important?
• What are the assumptions for logistic and linear regression?
• If you can build a perfect (100% accuracy) classification model to predict some customer behavior, what will be the problem in application?
• The probability that item an item at location A is 0.6 , and 0.8 at location B. What is the probability that item would be found on Amazon website?
• Given a ‘csv’ file with ID and Quantity columns, 50million records and size of data as 2 GBs, write a program in any language of your choice to aggregate the QUANTITY column.
• Implement circular queue using an array.
• When you have a time series data by monthly, it has large data records, how will you find out significant difference between this month and previous months values?
• Compare Lasso and Ridge Regression.
• What’s the difference between MLE and MAP inference?
• Given a function with inputs — an array with N randomly sorted numbers, and an int K, return output in an array with the K largest numbers.
• When users are navigating through the Amazon website, they are performing several actions. What is the best way to model if their next action would be a purchase?
• Estimate the disease probability in one city given the probability is very low national wide. Randomly asked 1000 person in this city, with all negative response(NO disease). What is the probability of disease in this city?
• Describe SVM.
• How does K-means work? What kind of distance metric would you choose? What if different features have different dynamic range?
• What is boosting?
Facebook Facebook经过近十年来累积大量数据之后，2013 年起，Facebook 内的工程师开始尝试使用 CNN。之后，Facebook 认识到 AI 和 Deep Learning 的重要性，并聘用了他们的第一位 AI 工程师——Google 大脑 Marc’Aurelio Ranzato。随后又聘请了 CNN 的发明人 Yann LeCun（现已不再负责 Facebook AI 研究院的领导工作）。

Facebook的面试流程和大多数公司一样，电面＋现场面试。

Facebook AI的阅读准备材料

1. Facebook’s AI framework for vision applications: Open Neural Network Exchange Format. (ONNX)

2. Download Library of their projects/packages: Downloads

3. Facebook Research Blog

• There is a building with 100 floors. You are given 2 identical eggs. How do you use 2 eggs to find the threshold floor, where the egg will definitely break from any floor above floor N, including floor N itself.
• You randomly draw a coin from 100 coins — 1 unfair coin (head-head), 99 fair coins (head-tail) and roll it 10 times. If the result is 10 heads, what is the probability that the coin is unfair?
• Write a sorting algorithm for a numerical dataset in Python.
• Facebook would like to develop a way to estimate the month and day of people’s birthdays, regardless of whether people give us that information directly. What methods would you propose, and data would you use, to help with that task?
• Use a python built-in package to manipulate ‘csv’ data.
• How would you compare the relative performance of two different back-end engines for automated generation of Facebook “Friend” suggestions?
• Given a KPI, choose the right metric, perform ETL. (using SQL/Code)
• You’re about to get on a plane to Seattle. You want to know if you should bring an umbrella. You call 3 random friends of yours who live there and ask each independently if it’s raining. Each of your friends has a 2/3 chance of telling you the truth and a 1/3 chance of messing with you by lying. All 3 friends tell you that “Yes” it is raining. What is the probability that it’s actually raining in Seattle?
• Consider a game with 2 players, A and B. Player A has 8 stones, player B has 6. Game proceeds as follows. First, A rolls a fair 6-sided die, and the number on the die determines how many stones A takes over from B. Next, B rolls the same die, and the exact same thing happens in reverse. This concludes the round. Whoever has more stones at the end of the round wins and the game is over. If players end up with equal # of stones at the end of the round, it is a tie and another round ensues. What is the probability that B wins in 1, 2, …, n rounds?
• How do you get the count of each letter in a sentence?
• How do you prove that males are on average taller than females by knowing just gender or height?
• What is a monkey patch?
• Given a list A of objects and another list B which is identical to A except that one element is removed, find the removed element.
• Given a list of integers (positive & negative), write an algorithm to find whether there’s at least a pair of integers that sum up to zero. How would you improve your algorithm’s performance?
• Make a histogram of 2 variables.
• Build a histogram of post reply count in SQL (number of posts with x replies, x+1 replies, etc).
• Build a table with a summary of feature usage per user every day (keep track of the last action by users and roll up every day).
• You’re at a casino with two dice, if you roll a 5 you win, and get paid \$10. What is your expected payout? If you play until you win (however long that takes) then stop, what is your expected payout?
• What metric would you show small businesses if you were trying to have them sign up for Facebook Ads?
• Given a table of friend requests sent and friend requests received, find the user with the most friends.
• Likes/user and minutes spent on a platform are increasing but total number of users are decreasing. What could be the root cause of it?
• How many high schools that people have listed on their profiles are real? How do we find out, and deploy at scale, a way of finding invalid schools?
• How do you map nicknames (Pete, Andy, Nick, Rob, etc) to real names?
• Facebook sees that likes are up 10% year over year, why could this be?
• If a PM says that they want to double the number of ads in Newsfeed, how would you figure out if this is a good idea or not?

ref: https://medium.com/@vimarshk   VIP定制项目 