About Kantar : Kantar is a data and evidence-based agency providing insights and actionable recommendations to clients, worldwide. We have a complete, unique and rounded understanding of people around the world: how they think, feel and act, globally and locally in over 90 markets. We don’t just help clients understand what’s happened, we tell them why and how they can shape the future.
Interviewing for Role : Data Science Innovation – I
Key Skills Required : Inferential Statistics, Linear algebra, Predictive Modelling algorithms (parametric, tree-based, boosting), Unsupervised Learning algorithms, Deep Learning, Image Processing, CNNs, RNNs, Python/R
Round 1 | Telephonic Interview by Business Head
Started with an introduction to Kantar, how diverse its business is and which division I’ll be working on. Proceeded to “Tell me something about yourself” and finally about the projects mentioned in my resume. Wouldn’t consider this as an interview, rather a smooth conversation where both parties got to know about each other.
Things we discussed:
(1) Mostly background and experience
(2) Projects mentioned on resume
(3) If I’m open to relocation
Round 2 | Telephonic Technical Interview by AI Team Lead
Grilling interview by the Team Lead where we discussed mostly mathematical concepts. Went on for an hour at the end of which he told that the HR would get in touch regarding on-site interview schedule (Chennai)
(a) Started with overview of projects on resume
(b) Asked me in detail what steps were taken for the Image classification project starting from data acquisition to evaluation.
(c) Basic questions on statistical tests (t-test, hypothesis tests etc.), i.e. their assumptions, applications and limitations.
(d) Backpropagation and how it works, Perceptron vs Linear networks, choice of activation functions and evaluation metrics
(e) Advantages of CNNs over classical image processing techniques, usage of filters, strides, padding
(f) Optimisation techniques (gradient descent, SGD, Adadelta, Adam)
Round 3 | On-site Interview by R&D Head
Taken in Chennai at their office premises where I was exposed to their working environment and met with the team.
(a) Started with projects and interviewer was mostly interested in the challenges associated with data cleaning and how it was dealt in the Image Classification project
(b) Why was ResNet used and not any other DL/ML method? Could it have been better? Etc. questions were asked on the same topic
(c) Asked a couple of questions on effectiveness of boosting (especially gradient boosting) especially to unbalanced classification scenarios
Round 4 | On-site Interview by Team 1 (Data Science Innovation – II)
Taken on the same day in the second half and was purely technical.
(a) Was asked to draw the architecture of ResNet, why skip connections were used and vanishing and exploding gradients
(b) Loss function used and derivation of weight updates
Round 5 | On-site Interview by Team 2 (Data Science Innovation – III)
Taken by a senior team in the second half and was purely technical.
(a) Favorite topic- bias-variance trade-off, what it is, its interpretation and formula. Drew a picture and asked to identify high bias, low bias incidents
(b) Assumption of Naive Bayes algorithm, why it’s so widely used despite shortcomings
(c) Assumption of linear regression, evaluation of best-fit line, OLS vs gradient descent, independent variable treatment for applying linear regression
Final Outcome : SELECT
What I think worked for me:
Having a sound mathematical foundation really helped in getting through the initial rounds. In the later rounds, my working knowledge on Deep Learning and image algorithms helped. Also, asking the right questions to the interviewer helped me display eagerness to learn on the job.