What Is the Unit Rate if Kenny Reads 5/8 of a Page in 2/3 of a Minute

Introduction

Back in 2009, deep learning was only an emerging field. Only a few people recognised information technology equally a fruitful area of research. Today, it is existence used for developing applications which were considered difficult or incommunicable to exercise till some time back.

Speech recognition, image recognition, finding patterns in a dataset, object classification in photographs, character text generation, self-driving cars and many more than are just a few examples. Hence information technology is important to exist familiar with deep learning and its concepts.

In this skilltest, we tested our community on basic concepts of Deep Learning. A total of 1070 people participated in this skill test.

If you missed taking the test, here is your opportunity to wait at the questions and check your skill level. If you are just getting started with Deep Learning, here is a grade to assist y'all in your journey to Primary Deep Learning:

• Certified AI & ML Blackbelt+ Program

Overall Results

Beneath is the distribution of scores, this volition help you evaluate your performance:

You can access your performance here. More 200 people participated in the skill test and the highest score was 35. Hither are a few statistics about the distribution.

Overall distribution

Hateful Score: 16.45

Median Score: twenty

Mode Score: 0

Information technology seems like a lot of people started the competition very tardily or didn't take it beyond a few questions. I am not completely sure why, but may exist because the field of study is avant-garde for a lot of audience.

If you take any insight on why this is so, do let us know.

Helpful Resources

Fundamentals of Deep Learning – Starting with Artificial Neural Network

Practical Guide to implementing Neural Networks in Python (using Theano)

A Complete Guide on Getting Started with Deep Learning in Python

Tutorial: Optimizing Neural Networks using Keras (with Image recognition case study)

An Introduction to Implementing Neural Networks using TensorFlow

Questions and Answers

Q1. A neural network model is said to be inspired from the human brain.

The neural network consists of many neurons, each neuron takes an input, processes it and gives an output. Here's a diagrammatic representation of a real neuron.

Which of the following statement(due south) correctly represents a real neuron?

A. A neuron has a single input and a single output only

B. A neuron has multiple inputs but a single output merely

C. A neuron has a unmarried input but multiple outputs

D. A neuron has multiple inputs and multiple outputs

E. All of the above statements are valid

Solution: (Eastward)

A neuron tin can take a single Input / Output or multiple Inputs / Outputs.

Q2. Beneath is a mathematical representation of a neuron.

The unlike components of the neuron are denoted as:

• x1, x2,…, xN: These are inputs to the neuron. These tin either be the actual observations from input layer or an intermediate value from one of the hidden layers.
• w1, w2,…,wN: The Weight of each input.
• bi: Is termed as Bias units. These are constant values added to the input of the activation function corresponding to each weight. Information technology works similar to an intercept term.
• a:  Is termed as the activation of the neuron which tin can exist represented as
• and y: is the output of the neuron

Considering the in a higher place notations, will a line equation (y = mx + c) fall into the category of a neuron?

A. Yep

B. No

Solution: (A)

A unmarried neuron with no non-linearity can exist considered as a linear regression role.

Q3. Let u.s.a. assume we implement an AND function to a single neuron. Below is a tabular representation of an AND part:

X1	X2	X1 AND X2
0	0	0
0	1	0
one	0	0
ane	1	1

The activation function of our neuron is denoted as:

What would exist the weights and bias?

(Hint: For which values of w1, w2 and b does our neuron implement an AND office?)

A. Bias = -1.5, w1 = 1, w2 = 1

B. Bias = i.5, w1 = ii, w2 = two

C. Bias = 1, w1 = 1.5, w2 = 1.5

D. None of these

Solution: (A)

A.

1. f(-1.5*i + one*0 + 1*0) = f(-i.5) = 0
2. f(-one.five*1 + 1*0 + 1*i) = f(-0.5) = 0
3. f(-1.5*1 + 1*ane + 1*0) = f(-0.five) = 0
4. f(-1.5*1 + 1*ane+ one*1) = f(0.v) = 1

Therefore choice A is correct

Q4. A network is created when nosotros multiple neurons stack together. Allow usa take an example of a neural network simulating an XNOR function.

You can encounter that the concluding neuron takes input from 2 neurons earlier information technology. The activation function for all the neurons is given by:

 Suppose X1 is 0 and X2 is i, what will be the output for the above neural network?

A. 0

B. one

Solution: (A)

Output of a1: f(0.5*1 + -1*0 + -1*1) = f(-0.5) = 0

Output of a2: f(-one.5*1 + one*0 + i*1) = f(-0.5) = 0

Output of a3: f(-0.v*i + 1*0 + 1*0) = f(-0.5) = 0

So the right reply is A

Q5. In a neural network, knowing the weight and bias of each neuron is the most important stride. If you tin can somehow get the correct value of weight and bias for each neuron, yous can gauge any function. What would exist the best mode to approach this?

A. Assign random values and pray to God they are correct

B. Search every possible combination of weights and biases till you get the best value

C. Iteratively check that subsequently assigning a value how far y'all are from the best values, and slightly modify the assigned values values to make them amend

D. None of these

Solution: (C)

Option C is the description of gradient descent.

Q6. What are the steps for using a gradient descent algorithm?

1. Summate fault between the bodily value and the predicted value
2. Reiterate until y'all find the best weights of network
3. Laissez passer an input through the network and get values from output layer
4. Initialize random weight and bias
5. Go to each neurons which contributes to the error and alter its respective values to reduce the fault

A. 1, 2, 3, 4, v

B. 5, 4, iii, 2, 1

C. 3, two, 1, 5, iv

D. 4, 3, ane, 5, 2

Solution: (D)

Option D is correct

Q7. Suppose y'all have inputs every bit x, y, and z with values -2, five, and -4 respectively. You lot accept a neuron 'q' and neuron 'f' with functions:

q = x + y

f = q * z

Graphical representation of the functions is as follows:

 What is the slope of F with respect to ten, y, and z?

(HINT: To calculate slope, you must find (df/dx), (df/dy) and (df/dz))

A. (-iii,iv,4)

B. (4,iv,3)

C. (-four,-4,iii)

D. (3,-4,-four)

Solution: (C)

Choice C is correct.

Q8. Now let's revise the previous slides. We have learned that:

• A neural network is a (crude) mathematical representation of a brain, which consists of smaller components called neurons.
• Each neuron has an input, a processing function, and an output.
• These neurons are stacked together to form a network, which tin be used to approximate whatever function.
• To get the best possible neural network, we can utilise techniques like gradient descent to update our neural network model.

Given above is a description of a neural network. When does a neural network model get a deep learning model?

A. When you add more subconscious layers and increment depth of neural network

B. When there is higher dimensionality of information

C. When the problem is an image recognition problem

D. None of these

Solution: (A)

More depth ways the network is deeper. There is no strict rule of how many layers are necessary to make a model deep, but all the same if in that location are more than two hidden layers, the model is said to be deep.

Q9. A neural network can be considered as multiple simple equations stacked together. Suppose we want to replicate the function for the below mentioned decision boundary.

Using two simple inputs h1 and h2

What will be the final equation?

A. (h1 AND Not h2) OR (Not h1 AND h2)

B. (h1 OR NOT h2) AND (NOT h1 OR h2)

C. (h1 AND h2) OR (h1 OR h2)

D. None of these

Solution: (A)

As you lot can see, combining h1 and h2 in an intelligent way can get you a complex equation easily. Refer Chapter 9 of this book

Q10. "Convolutional Neural Networks can perform diverse types of transformation (rotations or scaling) in an input". Is the argument correct True or Imitation?

A. True

B. False

Solution: (B)

Data Preprocessing steps (viz rotation, scaling) is necessary before y'all give the data to neural network because neural network cannot do information technology itself.

Q11. Which of the post-obit techniques perform like operations every bit dropout in a neural network?

A. Bagging

B. Boosting

C. Stacking

D. None of these

Solution: (A)

Dropout can be seen as an farthermost form of bagging in which each model is trained on a single instance and each parameter of the model is very strongly regularized by sharing information technology with the corresponding parameter in all the other models. Refer here

Q 12. Which of the following gives not-linearity to a neural network?

A. Stochastic Gradient Descent

B. Rectified Linear Unit

C. Convolution function

D. None of the to a higher place

Solution: (B)

Rectified Linear unit is a non-linear activation function.

Q13. In training a neural network, yous notice that the loss does not decrease in the few starting epochs.

The reasons for this could be:

1. The learning is rate is depression
2. Regularization parameter is loftier
3. Stuck at local minima

What according to you lot are the probable reasons?

A. 1 and 2

B. 2 and three

C. 1 and 3

D. Whatever of these

Solution: (D)

The problem can occur due to whatsoever of the reasons mentioned.

Q14. Which of the following is true about model chapters (where model capacity ways the ability of neural network to approximate complex functions) ?

A. As number of hidden layers increase, model chapters increases

B. Equally dropout ratio increases, model capacity increases

C. Equally learning charge per unit increases, model capacity increases

D. None of these

Solution: (A)

Only option A is right.

Q15. If you increase the number of hidden layers in a Multi Layer Perceptron, the classification fault of test data always decreases. True or False?

A. True

B. False

Solution: (B)

This is not always true. Overfitting may cause the error to increase.

Q16.You are edifice a neural network where it gets input from the previous layer as well as from itself.

Which of the following architecture has feedback connections?

A. Recurrent Neural network

B. Convolutional Neural Network

C. Restricted Boltzmann Motorcar

D. None of these

Solution: (A)

Option A is correct.

Q17. What is the sequence of the following tasks in a perceptron?

1. Initialize weights of perceptron randomly
2. Get to the next batch of dataset
3. If the prediction does not match the output, change the weights
4. For a sample input, compute an output

A. ane, 2, three, 4

B. 4, iii, 2, 1

C. 3, 1, 2, iv

D. 1, 4, three, two

Solution: (D)

Sequence D is correct.

Q18. Suppose that you have to minimize the toll function by changing the parameters. Which of the following technique could be used for this?

A. Exhaustive Search

B. Random Search

C. Bayesian Optimization

D. Any of these

Solution: (D)

Any of the in a higher place mentioned technique can be used to change parameters.

Q19. Showtime Club Gradient descent would not piece of work correctly (i.e. may go stuck) in which of the following graphs?

A.

B.

C.

D. None of these

Solution: (B)

This is a archetype example of saddle indicate trouble of gradient descent.

Q20. The beneath graph shows the accuracy of a trained 3-layer convolutional neural network vs the number of parameters (i.e. number of feature kernels).

The tendency suggests that as you increment the width of a neural network, the accuracy increases till a certain threshold value, and so starts decreasing.

What could be the possible reason for this decrease?

A. Even if number of kernels increase, only few of them are used for prediction

B. As the number of kernels increase, the predictive power of neural network decrease

C. As the number of kernels increase, they offset to correlate with each other which in turn helps overfitting

D. None of these

Solution: (C)

As mentioned in option C, the possible reason could be kernel correlation.

Q21. Suppose nosotros have ane subconscious layer neural network every bit shown above. The hidden layer in this network works every bit a dimensionality reductor. Now instead of using this hidden layer, we supervene upon information technology with a dimensionality reduction technique such as PCA.

Would the network that uses a dimensionality reduction technique always requite same output as network with hidden layer?

A. Yes

B. No

Solution: (B)

Because PCA works on correlated features, whereas hidden layers work on predictive capacity of features.

Q22. Tin can a neural network model the function (y=1/ten)?

A. Yes

B. No

Solution: (A)

Pick A is true, considering activation part tin be reciprocal function.

Q23. In which neural internet architecture, does weight sharing occur?

A. Convolutional neural Network

B. Recurrent Neural Network

C. Fully Connected Neural Network

D. Both A and B

Solution: (D)

Option D is correct.

Q24. Batch Normalization is helpful because

A. It normalizes (changes) all the input earlier sending it to the next layer

B. Information technology returns back the normalized mean and standard deviation of weights

C. It is a very efficient backpropagation technique

D. None of these

Solution: (A)

To read more most batch normalization, come across refer this video

Q25. Instead of trying to achieve absolute zip mistake, we set up a metric called bayes fault which is the fault we promise to achieve. What could exist the reason for using bayes error?

A. Input variables may not contain consummate information virtually the output variable

B. Organisation (that creates input-output mapping) may be stochastic

C. Limited training information

D. All the above

Solution: (D)

In reality achieving accurate prediction is a myth. So we should promise to achieve an "achievable result".

Q26. The number of neurons in the output layer should friction match the number of classes (Where the number of classes is greater than 2) in a supervised learning task. Truthful or Fake?

A. True

B. False

Solution: (B)

It depends on output encoding. If it is one-hot encoding, then its true. But you can have two outputs for four classes, and take the binary values as four classes(00,01,10,11).

Q27. In a neural network, which of the post-obit techniques is used to deal with overfitting?

A. Dropout

B. Regularization

C. Batch Normalization

D. All of these

Solution: (D)

All of the techniques can be used to deal with overfitting.

Q28. Y = ax^two + bx + c (polynomial equation of degree two)

Can this equation be represented by a neural network of single hidden layer with linear threshold?

A. Aye

B. No

Solution: (B)

The reply is no because having a linear threshold restricts your neural network and in simple terms, makes it a consequential linear transformation part.

Q29. What is a dead unit in a neural network?

A. A unit of measurement which doesn't update during grooming by any of its neighbour

B. A unit which does not respond completely to any of the training patterns

C. The unit which produces the biggest sum-squared mistake

D. None of these

Solution: (A)

Option A is correct.

Q30. Which of the following statement is the best description of early stopping?

A. Train the network until a local minimum in the error part is reached

B. Simulate the network on a test dataset after every epoch of training. Finish preparation when the generalization error starts to increment

C. Add a momentum term to the weight update in the Generalized Delta Dominion, and then that training converges more chop-chop

D. A faster version of backpropagation, such equally the `Quickprop' algorithm

Solution: (B)

Option B is right.

Q31. What if we use a learning rate that's likewise large?

A. Network will converge

B. Network will non converge

C. Can't Say

Solution: B

Choice B is correct because the mistake rate would become erratic and explode.

Q32. The network shown in Figure 1 is trained to recognize the characters H and T as shown below:

What would be the output of the network?

1. 
2. 
3. 
4. Could exist A or B depending on the weights of neural network

Solution: (D)

Without knowing what are the weights and biases of a neural network, we cannot comment on what output information technology would give.

Q33. Suppose a convolutional neural network is trained on ImageNet dataset (Object recognition dataset). This trained model is then given a completely white image every bit an input.The output probabilities for this input would be equal for all classes. True or False?

A. True

B. False

Solution: (B)

There would be some neurons which are do not activate for white pixels as input. Then the classes wont be equal.

Q34. When pooling layer is added in a convolutional neural network, translation in-variance is preserved. True or False?

A. True

B. False

Solution: (A)

Translation invariance is induced when you use pooling.

Q35. Which gradient technique is more advantageous when the information is too big to handle in RAM simultaneously?

A. Full Batch Gradient Descent

B. Stochastic Gradient Descent

Solution: (B)

Option B is right.

Q36. The graph represents gradient menstruum of a iv-hidden layer neural network which is trained using sigmoid activation function per epoch of grooming. The neural network suffers with the vanishing gradient problem.

Which of the following statements is truthful?

A. Hidden layer 1 corresponds to D, Subconscious layer two corresponds to C, Hidden layer 3 corresponds to B and Hidden layer 4 corresponds to A

B. Subconscious layer ane corresponds to A, Subconscious layer 2 corresponds to B, Hidden layer 3 corresponds to C and Subconscious layer 4 corresponds to D

Solution: (A)

This is a description of a vanishing gradient trouble. As the backprop algorithm goes to starting layers, learning decreases.

Q37. For a classification task, instead of random weight initializations in a neural network, we prepare all the weights to zero. Which of the following statements is truthful?

A. There will non be any problem and the neural network will railroad train properly

B. The neural network will train but all the neurons will end upwards recognizing the aforementioned thing

C. The neural network will not train equally at that place is no internet slope alter

D. None of these

Solution: (B)

Choice B is right.

Q38. There is a plateau at the kickoff. This is happening because the neural network gets stuck at local minima before going on to global minima.

To avoid this, which of the following strategy should work?

A. Increase the number of parameters, as the network would not go stuck at local minima

B. Decrease the learning charge per unit by ten times at the start and and then use momentum

C. Jitter the learning rate, i.e. change the learning rate for a few epochs

D. None of these

Solution: (C)

Option C can be used to take a neural network out of local minima in which it is stuck.

Q39. For an prototype recognition trouble (recognizing a cat in a photograph), which architecture of neural network would exist better suited to solve the problem?

A. Multi Layer Perceptron

B. Convolutional Neural Network

C. Recurrent Neural network

D. Perceptron

Solution: (B)

Convolutional Neural Network would exist ameliorate suited for image related bug because of its inherent nature for taking into business relationship changes in nearby locations of an epitome

Q40.Suppose while grooming, you encounter this effect. The error all of a sudden increases after a couple of iterations.

Yous determine that there must a problem with the data. You lot plot the data and discover the insight that, original data is somewhat skewed and that may be causing the problem.

What volition yous practice to deal with this claiming?

A. Normalize

B. Apply PCA then Normalize

C. Take Log Transform of the information

D. None of these

Solution: (B)

First you lot would remove the correlations of the data and so zero center it.

Q41. Which of the following is a decision boundary of Neural Network?

A) B

B) A

C) D

D) C

E) All of these

Solution: (E)

A neural network is said to be a universal office approximator, and then it can theoretically correspond whatsoever decision boundary.

Q42. In the graph below, nosotros find that the fault has many "ups and downs"

Should nosotros be worried?

A. Aye, because this ways there is a problem with the learning rate of neural network.

B. No, equally long every bit there is a cumulative decrease in both training and validation error, nosotros don't need to worry.

Solution: (B)

Option B is right. In order to decrease these "ups and downs" try to increase the batch size.

Q43. What are the factors to select the depth of neural network?

1. Type of neural network (eg. MLP, CNN etc)
2. Input data
3. Computation power, i.east. Hardware capabilities and software capabilities
4. Learning Charge per unit
5. The output role to map

A. 1, 2, 4, five

B. 2, 3, 4, five

C. 1, iii, 4, five

D. All of these

Solution: (D)

All of the higher up factors are important to select the depth of neural network

Q44. Consider the scenario. The problem you are trying to solve has a small amount of information. Fortunately, you have a pre-trained neural network that was trained on a similar problem. Which of the post-obit methodologies would you choose to make utilise of this pre-trained network?

A. Re-train the model for the new dataset

B. Assess on every layer how the model performs and only select a few of them

C. Fine melody the terminal couple of layers merely

D. Freeze all the layers except the final, re-train the terminal layer

Solution: (D)

If the dataset is mostly similar, the best method would exist to train only the concluding layer, as previous all layers piece of work as feature extractors.

Q45. Increase in size of a convolutional kernel would necessarily increase the operation of a convolutional network.

A. True

B. Fake

Solution: (B)

Increasing kernel size would not necessarily increment performance. This depends heavily on the dataset.

End Notes

I hope you lot enjoyed taking the test and you lot found the solutions helpful. The test focused on conceptual cognition of Deep Learning.

We tried to clear all your doubts through this article but if we take missed out on something and then permit me know in comments below. If you accept any suggestions or improvements you think nosotros should brand in the adjacent skilltest, let u.s. know by dropping your feedback in the comments section.

Learn, compete, hack and get hired!

menardwineve.blogspot.com

Source: https://www.analyticsvidhya.com/blog/2017/01/must-know-questions-deep-learning/

Share this post