We took a 3-hour test on May 2 which was conducted via remote proctoring. The test had 2 subjects extra besides the GATE syllabus – computer graphics and AI. The results were posted on IITB CSE dept’s portal the next day and while theory and computing systems got lesser than the max possible intake, intelligent systems received a lot of applications and they decided to shortlist more candidates in IS than their earlier limit, probably to make up for lesser candidates in the other two categories.

I was shortlisted for RAP and TA. My first interview was with the Cryo-electron microscopy project on May 10, and it was a disaster.


Prof Ajit took the interview. He asked me to give an introduction, what I’d been doing for the past few years after graduation, why I had selected the CEM project (I could not give a good answer), and then finally asked me a linear regression problem. At this point, I had covered only 15 lectures of Gilbert Strang’s LA course, and while I knew about least squares and projection, I could not give the satisfactory ML-style answer that he was looking for. In 10 mins it was over.


Profs PB and PC took this interview on May 11. Nothing could have prepared me for it.

PC took my introduction, and then PB decided to start. They had actually read my SOP, or had it open at that moment, and asked me about my interest in graph mining. He then said let’s start with graphs (I had prepared basic LA and probability), and I knew I was a goner.

PB: What are Euler and Hamiltonian graphs?
Me: Answered

PB: Which problem is the harder of the two?

Me: Explained that Euler circuits can be checked by checking the degrees of the nodes in the graph. Hamiltonian was a known hard problem.

PB: Explain formally what you mean by “hard”.

Me: Tried to explain the concept of reductions but I could not remember the correct direction of the reduction (from or to known NP complete problems).

PB: Asked the time complexity of the reduction.

Me: Could not answer this one.

PB: Said that I should know the correct direction of the reduction, and that many get it wrong. Then decided to ask probability. Asked about the basic axioms of probability and what Bayes’ rule is and what it means.

Me: Explained these.

PB: Asked me about Naive Bayes classifiers and framed a Bayesian problem (“patient has blood vitals, you can consider them probabilities – we want to find whether he has cancer. How will you frame the equation”)

Me: Couldn’t do it because I had no idea how to write naive Bayes stuff.

I can’t remember what else PB asked me, but right after that, he said he was done in an authoritative tone (if he had been any more authoritative, I would probably have exited my own house chup chaap, forget about leaving the meeting).

PC started asking about linear algebra. I felt like I still had a chance.

PC: Asked about what orthonormal vectors are.

Me: Explained.

PC: Asked what the main property of an orthogonal matrix is.

Me: Got stumped. It’s $Q^{-1} = Q^T$. He then showed it through row-column dot product of the vectors during matrix multiplication. Safe to say I no longer felt like I had a chance.

PC: Shifted discussion to eigenvalues and eigenvectors. Asked me about what they are.

Me: Explained, although not straight to the point (should have just talked about $Ax = \lambda.x$ instead of saying $x$ is a vector in the nullspace of $A – \lambda.I$).

PC: Asked about geometrical interpretation.

Me: Explained about how it produces a new vector parallel to the earlier one. He said “scaled up/down versions of it” and I agreed.

PC: Asked me about eigenvalues of the inverse of the matrix. Asked me to prove my answer.

Me: Gladly proved how they are $1/\lambda$. Probably the only topic I got right in full.

PC: Asks what happens if the magnitude of eigenvalues is less than 1 and they are repeatedly multiplied with the vector.

Me: Obviously the term vanishes as we approach infinity. (this was probably about the diagonal equation and powers of a matrix – I should have said that n independent eigenvectors are required).

PC: Suddenly starts joking about whether I have lost touch with CS after so many years.

Me: I didn’t even know what to say to that.

Suddenly he stopped the teasing (it looked like he saw something on the screen) and asked whether my GATE score was of 2021. I said yes. He was serious for a few moments.

Then he starts joking around again that I should be careful with my answers. Asks me what programming languages I have worked in – I said python and JS.

PC: What kind of languages are these?

Me: At this point I was nervous enough that I said typeless – but quickly corrected to dynamically typed and interpreted.

PC: Is type checking done in Python and JS, as opposed to a language like Java?
Me: Tried to explain that Java is a compiled and bytecode-interpreted language and during compilation, it has a semantic analysis phase where types are checked. But Python/JS wouldn’t have that. Maybe the CPython interpreter would but not at the Python level.

PC: Tried to trick me on this, asked which is faster.

Me: Mistakenly ended up saying that the interpreted ones are faster – yep, an experienced guy saying this.

He had a big smile and told me to be careful before answering.

At this point they stopped the interview and I thanked them and exit. Took roughly 30 mins.

posted in Interview Experience May 25, 2021