In a research scientist interview, you will be expected to show that you have the necessary technical knowledge and expertise pertaining to the specific position you are applying for. Some of the common topics include basic statistical methods, machine learning concepts and case study analysis. Also, the interviewer will most probably assess your communication and interpersonal skills, which are essential for effective teamwork and funding acquisition.
Here are three of the top research scientist interview questions and tips on how to answer them:
How to answer: Basically, such an interview question asks for a textbook recall of a certain machine learning concept and its conditions and applications. Avoid overcomplicating it. Just give a simple and straightforward answer that shows that you have a solid grasp of the concept.
How to answer: The interviewer wants to evaluate your problem-solving skills. Carefully choose a challenging situation that best reflects your ability to solve problems and explain what you did to overcome it. Preferably, the problem should be one that is relevant to your desired position.
How to answer: If you had successfully secured research funding in the past, you can talk about some of the methods you used. If not, highlight the abilities you possess that can help you acquire funding, such as grant writing skills and networking skills.
↳
For the questions 1: I think both options have the same expected value of 4 For the question 2: Use binomial distribution function. So basically, for one case to happen, you will use this function p(one case) = (0.96)^99*(0.04)^1 In total, there are 100 positions for the ad. 100 * p(one case) = 7.03% Less
↳
For "MockInterview dot co": The binomial part is correct but you argue that the expected value for option 2 is not 4 but this is false. In both cases E(x) = np = 100*(4/100) = 4 and E(x) = np=100*(1/25) = 4 again. Less
↳
Chance of getting exactly one add is ~7% As the formula is (NK) (0,04)^K * (0,96)^(N−K) where the first (NK) is the combination number N over K Less
↳
Since if two people become friends, the request has to be accepted. We may use the accepted table only for the question how many friends each id has. However, one person can either send or accept friend, we will need to remove the duplication. select a.accepter_id, count(*) as cnt from (select distinct accepter_id, send_id from accepted union select distinct send_id as accepter_id, accepter_id as send_id from accepted ) a group by accpeter_id order by cnt limit 1; Less
↳
In the vein of answers 6 and 7: SELECT a.user, COUNT(DISTINCT a.friend) AS friend_count FROM ( (SELECT accepter_id AS user, sender_id AS friend FROM ACCEPTED) UNION (SELECT sender_id AS user, acceptor_id AS friend FROM ACCEPTED) ) a GROUP BY a.user ORDER BY friend_count LIMIT 1; Less
↳
I think this would be simpler select requester_id from request_accepted union all select accepter_id from request_accepted) t group by 1 order by count desc limit 1 Less
↳
Good points above. A few more things to consider: - iOS vs Android demographic differences (age: Are iOS users younger on average and thus more likely to use Instagram?; geography: How do majority of iOS countries (US/Canada/Japan) differ from Android countries?) - Integration of Instagram within the FB app: Is it the same for both OS? - Android has different OEMs. Does this difference hold for all OEMs? Less
↳
This question is really asking whether OS is a discriminant variable if we want to predict Instagram usage. It's likely that OS is just a proxy of other factors, such as age, country, education, engagement and etc. A quick way to check this is to run a decision tree and check feature importance. If the tree doesn't split on OS, then we should investigate into the factors that actually matter. If OS turns out to be a significant feature, then we need to investigate the reasons. Other answers have provided some possibilities, such as user experience and SEO. Less
↳
a. We are only considering mobile usage? b. Check demographics of Android and iOS user i. Implement 2x2 contingency tables based on age, race, location, occupation cross-sectioned with mobile device to determine if there is a relationship between any mobile device and the other races c. Determine if there is a time pattern, has this always been the case i. Calculate the percent of Facebook Users on iOS that use Instagram and the percent of Facebook Users on Android that use Instagram every 3 months since Instagram was acquired by Facebook until now and determine if there is a consistent pattern. If the % was the same for both mobile devices to a point in time I will look for which new features or new developments that might have occurred at the time to see if this was the cause of the difference between mobile devices. I would also look at the d. What type of actions are being done on Facebook and Instagram by the different users (posting, scrolling, sharing, liking, sending messages, shopping, watching videos) i. Are the actions of people using Facebook the same between mobile devices? (% of posting, % of each type of posting, the % time scrolling, etc.) Determine if the actions of people on Instagram the same between the mobile device? Then determine is there is a difference between these actions. Less
↳
Did you get your interview feedback about rejection immediately ? How long it took for them to respond? Less
↳
WelI it took them few weeks to reply me. All the best your job hunt!
↳
Nope, no forms and rejection. I believe you have higher chances of getting in. Probably, your one of the top candidates. HR probably screen you now. Dont worry too much.. I will wish you luck working in the company and make a change for attaining that title. I got other offers already too. :) Less
↳
Over the last 7 days, how many users made more than 10 searches? WITH search_agg as( SELECT user_id, count(search_id) as count_search FROM search WHERE date>current_date -interval '7 days' GROUP BY user_id HAVING count_search>10) SELECT count (distinct user_id) FROM search_agg Less
↳
Investigation: avg. # clicks per search (YoY for seasonality) # clicks per event type (histogram) # clicks by weekday / weekends good / bad if avg. # clicks per search has increased it might be bad cause users don't find relevant content, I would have also check the avg. time per link, if it has decreased it means the user didn't find relevant content I will also check the ratio between reacted events (arrive / interested) out of total clicked events , if the ratio has increased it might be good cause people reacted as it's relevant for them Less
↳
select count(distinct user_id) from ( select user_id, count(search_id) as cnt_search_7days from searches where date > datediff(today() -7) having count(search_id) > 10 ) a Less
↳
I assume that the goal of the A/B was to monitor the impact on Comments, Likes and Time spent. The increase in Comments probably confounds the increase in time spent. (Does the increase in comments resulted in time spent or vice-versa) Lets examine what does increase in Comments mean for the business. It has increased user engagement in its most richest form. Probably more than likes. This would help Facebook profile users better by understanding users meaningfully from their comments. This opens up the possibility of better targeting ad campaigns (impressions) in the future. The increase in comments could potentially offset the loss in information from the reduction in likes. Assuming that the 1% increase in time spent is statistically significant, and the effect translates to roughly 50% of the users (2 billion total users) and an average user spent 1/2 hour on their app. It would mean half a billion hours more. Thats significant eyeballs and opportunity for improving revenue through impressions and click rate. The feature seems promising for implementation. Less
↳
I would keep the feature on, because we spend more "energy" when we are commenting a post. Also time spent on Facebook is quite a good metric. 1% more time exposed to ads, or to engage...for me it is a good trade off...5% comments and 1% +time spent worth more than like...for me comment is really much more important than like... Less
↳
Likes, comments, time spent all goes into a model that predicts a customer's lifetime value. See if the new levels increase or decrease CLV and make the product launch decision accordingly Less
↳
thank you for sharing your experience. any insight into why no offer? based on your answers, it seems like you had solid answers + good approach? Less
↳
Yes, I thought it quite well as well. However the only feedback I got that in both cases studies I had a problem with my a/b test design (I am not an expert in this area so I am not sure what kind of answer they were expecting), also I think I had a problem with 1 of the KPI I gave Less
↳
For part 1, Question 1: I would first want to ask some clarifying questions: Do we mean average posts per day per user? This would be: total # posts/total # users/time period What time period should we use to calculate the rate? Let's say we look over the past week (7 days) SELECT SUM(CASE WHEN action = 'post' THEN 1 ELSE 0 END)/COUNT(DISTINCT user_id)/7 FROM post_actions WHERE date BETWEEN CURRENT_DATE and DATE_ADD(CURRENT_DATE, -1) Part 1, question 2 Clarify: does the active table contain one unique record for each user every day? SELECT country, SUM(CASE WHEN b.action = 'post' THEN 1 ELSE 0 END)/COUNT(DISTINCT a.user_id) AS post_rate FROM active a LEFT JOIN post_actions b ON a.user_id = b.user_id AND a.date = b.date WHERE a.date=DATE_ADD(CURRENT_DATE, -1) AND a.active = 1 GROUP BY country Less
↳
import matplotlib.pyplot as plt %matplotlib inline from random import randint ddef rand5(): return randint(1,5) # generate random integers from 1 to num def rand(num): r=0 for i in range(num): r += rand5() return r % num +1 b= np.zeros(100000) for count in range(100000): b[count] = rand(7) _= plt.hist(b) # not prettified, but this is for test only Less
↳
how did you come up with X ^ (log(7) / log(5))?
↳
X ^ (log(7) / log(5))
↳
As 16p=x^3-1=(x-1)(x^2+x+1), we have p=(x-1)(x^2+x+1)/16. Notice x^2+x+1 must be an odd number, then x-1 must be a multiple of 16. Clearly x-1=16 is the only solution, otherwise, let x-1=16k (k>=2), then p=k*(x^2+x+1) which is not a prime number. Less
↳
A real number
↳
X = 17 is the only solution. First we can write X = 2k + 1 then use some simple number theory techniques to discuss k: k = 1,2,4,8, p The only feasible solution is k = 8 Less
↳
A trap. This question seems to ask you make a choice between two solution when in reality the answer is that you must be both of them. There is a million way to conciliate the 2 but you need to find a way to explain that in some circumstance you have to be a detail person, and some other you see the big picture. Show them that you are flexible. Less
↳
I agree with Nikko. It does depend on whether you are applying for a management position or a support position. Either way both the support and managemnet type should be able to comprehend the problems of the detail and be able to do the detail to achieve the big picture. It has been my experience that the big picture person is the idealist with little comprehension of what it will take to achieve the goal. The little picture (or pixel) person is is the true thinker and problem solver. It is a lot easier if both are thinking up-down and down-up. That is a good relationship. Less
↳
This is an important question. A big-picture person usually adopts a top down approach considering relatively high level aspects e.g., business viability and the extent, where else the technology would impact, mid term and long term potential future of the technology etc. However a detail person would rather adopt a bottom up approach by identifying the well defined technical challenges and their potential solutions, other similar problems and their strengths and limitations... Less