# PhD Interview Experience in Computer Science dept at IITs

The pic above I selected for header just to inform that CSE/AI interviews which I gave are mostly about maths only. *Happy Reading*

This blog is intended to help the aspirants who would like to pursue PhD in CSE/AI at Indian Institute of Technology(IITs) or IISc. I would be sharing with you my written test/interview experiences at following institutes:

- IISc Bangalore
- IIT Kanpur
- IIT Kharagpur
- IIT Delhi
- IIT Hyderabad

All the above interviews/test of the above institutes I have given for the Autumn semester of 2021–22. Most of the IITs have B.Tech as an eligibility criteria and I was pursuing M.Tech at the time of interview with a CGPA of above 8 and got shortlisted for test/interview in all of the above institutes.

Broad topics covered under the interview & tests are Linear Algebra, Probability, Basics of ML & other CS subjects. I will list down the questions asked under the topics subheadings below:

**Linear Algebra**

i. Define vector subspace

ii. Whether invertible matrices are subspace of square matrices

iii. Define basis of vector space

iv. Describe a method to solve linear equations of 10–20 variables

v. How to calculate similarity between two vectors from different dimensional spaces

**Probability**

i. If two events are independent, whether their complements will be independent too

ii. if A & B are independent, P(A|B) = 0.3 , what will be P(A|B`)

iii. How to calculate covariance of two normal distributions

**Algorithms**

i. Find cycle in a directed graph

ii. Draw a circle with minimum radius which encompasses given points in 2-D space

iii. Check whether a given string is substring of other string

**Machine Learning**

i. What is Attention

ii. How is Embedding learned

iii. Why Naive Bayes Naive

iv. What if dependence between features is known, How to Calculate Bayesian Probability

v. How LSTM is better than RNN

vi. Explain the gates of LSTM and what are the alternatives of LSTM

vii. If there a language with no rules like ‘ I ate Apple’ , ‘ Apple at I’ both same which architecture will be used for training on such language

viii.Explain N-gram modelling and how to calculate bi-gram probability

ix. Can CNN be used on texts, why is it used on images ?

x. Visually explain an encoder decoder model

xi. Consider a NN with M units in input layer & N units in output layer with no hidden layer, what will be number of trainable parameters

xii. How to reduce variance of LSTM based model

**Misc**

i. Explain the process transition states in OS

ii. Difference between RISC & CISC

iii. Explain various addressing modes in Computer Architecture

iv. Draw graph of xy=c

v. Given any 4 points , which type of quadrilateral formed from those points

As most of these answers are easily available on Google, I am not including them in this post. If required may be in second part of it. I would like to summarise my experience of facing these questions in various interviews and some suggestions in following points :

- If you don’t know, say no
- When they ask for topic preferences, make sure to bring them to the topic you are very much comfortable with
- When they say,
*Are you sure about your answer*, 90% of the cases your answer is wrong - Take your time, cross check your answer before putting across your answer
- Selection or rejection can be based on one right/wrong answer depending on who have asked this question in panel and how much influence he/she holds over the panel
- Don’t lose heart on missing out on a difficult question , just skip it and drive the panel to your comfort zone

I had enough rejections & some selections(IITK, IITKGP) too in my kitty, so I think the above suggestions would be very helpful. Comments welcome.