C3 AI Intern Data Scientist Interview Experience — ML Multiple Choice and Production Request Scheduling

Question 1: Vanishing Gradient
Which of the following is NOT a potential solution to the vanishing gradient problem in a deep neural network?
Adding skip connections to the neural network architecture
Using a different activation function like ReLU
Raising the learning rate
Adding Batch Normalization between layers

Question 2: Boosting Classification
For the classification problem, Boosting is widely used because:
Diversity can be boosted because all ensemble members are created in parallel
Residual error has been minimized, which reduces the margin distribution
It attempts to maximize the margins on the training data
None of the above

Question 3: Underfit GBoost
You have trained a gradient boosted model which you find is underfitting. What parameter can you consider decreasing?
The depth of the tree
The number of leaves
The proportion of objects used to fit the tree
The minimum amount of objects per leaf

Question 4: Multi-collinearity
Consider the following algorithms:

1. Linear regression without any regularization
2. Linear regression with L1 regularization
3. Decision tree
4. Logistic regression
5. Gradient boosted tree
   Which of these algorithms is NOT affected by multi-collinearity of data?
   2, 3
   1, 2, 4
   2, 3, 5
   3, 5
   1, 2, 3, 4, 5
   None

Question 5: Correlation
Which of the following is in the right order?
(Given 7 scatter plots 1 through 7 from strong positive correlation to strong negative correlation)

1. 1 < 2 < 3 < 4
2. 1 > 2 > 3 > 4
3. 7 < 6 < 5 < 4
4. 7 > 6 > 5 > 4
   Options:
   1 and 2
   1 and 3
   1 and 4
   2 and 3

Question 6: knn k
What should k of the k nearest neighbor be in order to maximize the leave one out CV (cross validation) accuracy?
1
3
5
Any K would result in the same result
None of the above

Question 7: R square
For the training set, what would happen to the $R^2$ if you add a completely random feature into a linear regression model?
R square would increase
R square would stay the same
R square would decrease
Both A and B can happen
Both A and C can happen
Both B and C can happen
All A, B, C can happen

Question 8: Six Sided Dice
Given a fair six sided die, we roll it repeatedly and sum the outcomes until the sum is 61 or higher. What is the most probable sum at the end of this game?
61
62
63
64
65
66

Question 9: Minimum Reject Request
A production line received many production requests, given as tuples of production start and end time,. However, no two requests can be processed at the same time. For example, out of two requests
[(15,30), (10,20)]
, only one can be processed by the production line. Suppose there is a list of requests
[(s1, e1), (s2, e2) ...]
. Find the minimum number of requests that have to be rejected from the production line.