TechVidvan

Top 67 Frequently Asked TensorFlow Interview Questions and Answers

Grasping a concept is one thing, and the ability to prove your learning is another. Interviews are the phase which checks your skills and validates your performance.

This article provides you with a handbook of important TensorFlow Interview questions and Answers to help you prepare well for your TensorFlow interview process and brush up the TensorFlow concepts.

In this series, you will get 150+ TensorFlow Interview Questions and Answers in 2 different parts, that covers

  • TensorFlow Interview Questions and Answers for Beginners
  • TensorFlow Interview Questions and Answers for Experienced 

Let’s look upon some basic TensorFlow Interview questions and Answers that play a major role in TensorFlow interview for Freshers.

But if you are already aware of basic TensorFlow Interview questions and Answers properly, then you can move directly to TensorFlow Interview questions and Answers for Experienced Professionals.

So let’s start with Basic TensorFlow Interview questions and Answers!!

 

TensorFlow Interview questions and Answers for Freshers

1. What is TensorFlow?

TensorFlow is an open-source framework developed by Google for machine learning. This library allows to create, execute the algorithms which are user-friendly and suitable for a system.

It allows eager execution through the visualization, which it provides with the help of TensorBoard. It eases the calculations, thus reducing the complexity of computations.

2. What are tensors?

Tensors are the representation of multi-dimensional arrays or vectors. It is the combination of digits, representing the data in its coded form. The representation of tensors is in the form of edges. The data stored in vectors have a shape, which also defines the dimensionality of a matrix.

The tensors represent scalars as well as vectors. A tensor is an object having three unique properties: Name, shape, and dtype. The graphs also contain the operations of a tensor.

Any object which is present in the model initially is called a feature vector. Graphs contain a feature vector that helps to map tensors.

3. How many types of Tensors are there?

There are three types of tensors, namely: constants, variables, and placeholders.

1. Constants: It represents the node that stores only one value, which does not change with time. The parameters in sync with constant are value, name=’Const’, shape=None, dtype=None. Its denotation is tf.constant.

2. Variables: It represents the different values at each run; hence it is named as a variable. It prints the most current value when executed. Its denotation is tf.variable.

3. Placeholders: The creation of nodes for these tensors takes place, but data assignment takes place later. The assignment takes place when the input depends on an external entity. Hence, this tensor requires the only data type and shape defined initially.

4. What are TensorFlow servables? Also, explain TensorFlow Serving.

It is the central unit in the architecture of TensorFlow Serving, which serves as objects used by the clients in the process of computations. It offers flexible size and granularity. It consists of one lookup table to a tuple of interference models.

Typical servables include the following:
a TensorFlow SavedModelBundle (tensorflow::Session)
a lookup table for embedding or vocabulary lookups

It improves the functionality and performance with the help of improvements:

  • streaming results
  • experimental APIs
  • asynchronous modes of operation

TensorFlow serving allows the system of TensorFlow, which we use for machine learning. It allows us to create, use, and execute the algorithms for the system a user wants to build. It extends its functionality with the help of collaborating environments such as TensorBoard.

5. Explain the advantages and disadvantages of TensorFlow.

Advantages:

  • Open-source Platform
  • Data visualization
  • Parallelism
  • Scalable
  • Compatible
  • Keras friendly
  • Graphical support
  • Support for OS

Disadvantages:

  • Symbolic Loops
  • Dependency
  • Frequent updates
  • Architectural limitations
  • Slow speed
  • GPU support
  • Support for Windows
  • inconsistent

6. What are the variables in TensorFlow?

The objects in TensorFlow which can change the value according to the change in the state are known as variables. It prints the recent most value at the time of execution. The life of a variable starts when tf.variable.initializer initializes it.

And the life of the variable ends with the collapse of a session. It provides the state of an object which changes with time hence holds the recent updated value. It requires initialization with a declaration to work with the model.

Frequently Asked TensorFlow Interview Questions and Answers for Beginners

7. What are TensorBoard visualizations?

TensorFlow allows the formulation of graphs for the models, which is the collection of tensors and nodes.

TensorBoard helps to perform the visualization through the graphs. TensorBoard is a collection of web services that help in debugging, eager execution, and a dynamic runtime. It’s a tool that eases the debugging with its functionality and is the sole method of visualization.

8. Explain the code.

model = tf.keras.Sequential ([
  tf.keras.layers.Dense(10, activation=tf.nn.relu, input_shape=(4,)),  
  tf.keras.layers.Dense(10, activation=tf.nn.relu),
  tf.keras.layers.Dense(3)
])

Keras API helps to model the complexity of a system and provide an easy way to perform operations. It passes the data through the ReLU layer, which is an activation layer. Keras passes the data through two Dense layers with 10 nodes each, and an output layer with 3 nodes representing our label predictions.

9. Explain TensorFlow loss and optimizer.

The computation of the difference between the predicted and actual value using a function is known as the loss function. The value of this function defines the amount of difference in values.

Hence at each run, the Gradient Function optimizer checks for the changes which can help to improve the model. With the help of optimizer, the loss reduces to the minimum and attains maximum accuracy.

10. Why and how normalization is achieved in TensorFlow?

The TensorFlow system expects all inputs in a dimension. But the dataset we work does not contain significant values. Hence the normalization of data is required. We perform the batch normalization of the data.

data = tf.nn.batch_norm_with_global_normalization()

11. What are estimators?

Estimators are the high-level API, which helps to reduce the code length for training, evaluation, and prediction. It is of two types: pre-made estimators and estimators (base class).

12. What is epoch?

It defines the total time frame of training the data. Often many epochs run to train the dataset of 10s of thousands of entries. Using many epochs allows us to make a better generalization for the system.

13. Name some applications of TensorFlow.

The areas in which TensorFlow usage allows to create systems are:

  • Image Recognition
  • Voice Recognition
  • Video Detection
  • Text-based applications

14. Why are NumPy and Matplotlib used with TensorFlow?

NumPy supports multi-dimensional arrays, matrices. Hence it deals with the data in the form of arrays.
Matplotlib handles all the graphical representation in TensorFlow. It supports the MATLAB interface.

15. Explain the following example?

for epoch in range(training_epochs):
for (x, y) in zip(train_X, train_Y):
sess.run(optimizer, feed_dict = {X:x, Y:y})

The initializer runs and feeds all the training data through the epochs.

16. What is the MNIST dataset?

MNIST (Modified National Institute of Standards and Technology database) is the handwriting recognition dataset that helps to perform visual image recognition.

The dataset contains the greyscale images of the 28X28 dimension. There are 60,000 images for the training and 10,000 images for testing purposes.

17. Difference between CNN and RNN.

CNN RNN
Convolutional Neural Network Recurrent Neural Network
Known as the feed-forward model For the series of data
Memoryless model Requires memory to store previous inputs
Cannot handle sequential data Can handle Sequential data
Used for Image recognition Used for Text recognition
Can handle fixed length of input/ output  Can handle arbitrary lengths of input/ output 
Feature compatibility is more Feature compatibility is less 
Handles permanent data Handles temporary data

Top TensorFlow Interview Questions and Answers

18. Explain backpropagation.

After the loss calculation, the loss is given to the model for improvising the model. At each step back, the change in weights takes place which helps to reduce the loss and increase the accuracy.

19. What is embedding?

The vector representation of words is also known as word embedding. It refers to representing the words in the vector format. It helps to find the meaning of symbols so that it can make predictions.

20. How training and testing are performed?

We build each model with the help of a dataset that is either created or imported.. The dataset is divided into two portions: training and testing. Training data contains most of the values so that system gets apprehended by all types of cases.

Training data contains very few values compared to training. It tests the system for unknown values for better learning.

21. What are the activation functions?

The function which imparts the non-linear property to the output of the summation function is the activation function. The expected output represents the probability hence it restrains the value between 0 and 1.

22. What are loaders?

It manages the life cycle of servable. The loader API enables an infrastructure that is independent of data or product use cases and algorithms. Then we standardize the loader’s API for loading and unloading of servable.

23. Which Dashboards are supported by TensorFlow?

TensorFlow supports the following Dashboards:

  • Scalar Dashboard
  • Histogram Dashboard
  • Distributer Dashboard
  • Image Dashboard
  • Projector
  • Graph Explorer
  • Audio Dashboard
  • Text Dashboard

24. How deep learning and TensorFlow are related?

Deep learning is the class of machine learning which uses neural networks for its working. Hence it is also known as neural networks. TensorFlow is used for the implementation of deep learning.

Deep learning allows the system to learn the working of a system with the help of neural networks. TensorFlow helps to build such networks.

25. Explain the graph. What do these terms represent: operations and tensors?

All the complex computations in TensorFlow are implemented using graphs. Graphs are the collection of operations represented using nodes. The graph consists of nodes and edges connecting nodes.
Operations: The nodes in the graph represent operations.
Tensors: The edges in the graph represent the edges that carry weight with it.

26. State the difference between the two lines.

x = tf.constant(35, name = 'x')
y = tf.Variable(x+5, name = 'y')
print(y)

The first line declares a constant tensor that has value 35, name as x. The second line declares a variable with parameters values as x+5, name as y.

27. What are the vectors in TensorFlow?

A group of values is represented as arrays. Each value has an index representation and allows each point to be individually identified through its index position. Vectors have a magnitude and a direction.

28. What are the scalars in TensorFlow?

A scalar represents the single value, which cannot be denoted as space, i.e, it does not have directions. An entity such as temperature is a scalar.

29. Explain the term TensorFlow API.

TensorFlow has many APIs which are used for the implementation and execution of it. TensorFlow supports the following APIs:

  • Python
  • JavaScript
  • C++
  • Java
  • Go
  • Swift (Early Release)
  • R
  • Ruby
  • C#

30. What is the ReLU layer?

Rectified Linear Unit Layer acts as an activation layer which activates the function having a value above a specific unit. It replaces the negative values in an image with zero, defining a linear relationship of the variable with the input. It makes the input invariant to noise; hence it is known as subsampling.

31. What is the Pooling layer?

The reduction in the size of an image takes place when data runs through the pooling layer. It aims at targeting the sections individually. It either finds the maximum value or an average of a section to reduce the size. This process is known as down-sampling.

32. What is the source?

Sources module provides the Servables, which provide one or many servable streams. For each stream, Source provides one Loader for each version. Servables can detect servables from random storage systems.

Sources manage the state of servables or versions. Sources come in use where there are delta updates between versions of servable.

33. Why is validation performed?

The input data we work on require the following points to be checked:

  • The improper data type
  • Out of range values
  • No value

During the training, a model prepares a schema that contains the values the model should produce. Validation is done to compare the actual output with the schema.

So now when you have seen TensorFlow Interview Questions and Answers for Beginners, let’s move to Advanced TensorFlow Interview Questions and Answers.

TensorFlow Interview Questions and Answers for Experienced

1. What is dimensionality reduction?

The reduction of variables or dimensionality from the data for reducing its size and not losing the essential information. This process is known as dimensionality reduction.

2. What is an underfit and overfit condition?

The underfit condition arises when the model is not able to classify data. The situation occurs when data is not enough, or the model is linear or nonlinear.
The overfit situation arises when we train the model with a large amount of data, which leads to processing the data with noise and less accuracy. The non-linear model is the reason for overfitting.

3. Explain the concept of the ROC curve.

The ROC (Receiver Operating Characteristics) curve is the important evaluation metric which checks if the model can classify different classes of the model. The value 1 shows that it has the proper ability of classification, and 0 shows no ability of classification, and 0.5 shows no capability of classification.

4. What is the F1 score?

F1 score is the test for performance, which considers false positives and false negatives. Recall and precision help to calculate the F1 score. The calculated harmonic mean with the use of recall and precision is the f1 score.

It ranges between 0 and 1 where 0 represents the worst value and 1 is the best. It is the result of the balance between precision and recall.

5. Explain the following example.

vocab_size = 10000
model = keras.Sequential()
model.add(keras.layers.Embedding(vocab_size, 16))
model.add(keras.layers.GlobalAveragePooling1D())
model.add(keras.layers.Dense(16, activation=tf.nn.relu))
model.add(keras.layers.Dense(1, activation=tf.nn.sigmoid))

The code is for word embedding in which data is set to vectors using the embedding layer. The pooling layer converts the vector to finite length. The next 2 layers are the activation functions namely the ReLU layer and sigmoid function.

6. What is the KNN algorithm?

It is the structured algorithms to classify the data to k nearest neighbor. It does not learn from the training data. Despite this, it works on the dataset when classification occurs. It is a non-parametric algorithm that does not make any assumptions on the dataset.

7. How is the learning rate significant?

Setting a precise learning rate is critical as it helps the model to configure in different ways. Setting a low learning rate helps to learn the model with optimal weights, and a higher learning rate helps to learn the model faster with low optimal weights.

8. What does the Dense layer do?

The dense layer performs the function of the hidden layer. It passes the operation through an activation function, weights the function matrix, and creates the bias vector with the help of activation, Kernel, and bias, respectively.

9. Explain the term TensorFlow Transform.

TensorFlow Transform is used to pre-process the data over the features in the dataset. It can normalize data, convert strings, float numbers to integers. The graph for the data helps in training and serving.

10. What are type I and type II errors?

On the axis of prediction and actual result, the error can arise of two types:

1. Type I error- The error which arises when the model results as true when it has not occurred actually. It is a false positive value.

2. Type II error- The error which arises when a false hypothesis results as true in the research. It is a false negative value.

11. What happens when data misses in a dataset?

The dataset missing the files or data can either delete or recover the data using:

  • Listwise Deletion
  • Average or multiple Imputation
  • Substitution

12. Explain the following code?

model.compile(optimizer=tf.train.AdamOptimizer(), 
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])

The code calculates the loss and optimizes its value using the optimizer and the unit cross entropy.

13. What is k-clustering?

The method to categorize the vectors in the k points according to their mean coordinate is known as k-clustering. It calculates k means to cluster the vectors. Initially, we randomly point out k means.

The location of a vector is found to its nearest mean, and the recalculation for mean takes place. The points are categorized after k means calculation.

Commonly asked TensorFlow Interview Questions and Answers

14. Explain Recall and Precision.

Precision and Recall are the performance metrics i.e., they give insight about the model performance.

  1. Precision: The quotient of true result to the actual result. It gives the percentage of true positive to the sum of a true positive and false positive.
  2. Recall: It is the quotient of true results to the predicted result. It gives the percentage of true positive to the sum of a true positive and false negative.

15. How is Naive Bayes theorem significant in TensorFlow?

Naive Bayes algorithm helps the models to define the probability distributions. The classification models, such as object classification etc. are built using Naive Bayes in TensorFlow.

16. What algorithms can we apply to TensorFlow?

  • Linear regression
  • Linear classification
  • Boosted tree classification
  • Boosted tree regression
  • Deep learning classification

17. Explain scope with an example.

It is the process of using the variables outside the scope with the help of the name scope. The variables defined inside name_scope help to reduce the complexity of a program and its operations.

def my_op(a, b, c, name=None):
with tf.name_scope("MyOp") as scope:
a = tf.convert_to_tensor(a, name="a")
b = tf.convert_to_tensor(b, name="b")
c = tf.convert_to_tensor(c, name="c")
# Define some computation that uses `a`, `b`, and `c`.
return foo_op(..., name=scope)

18. Explain PCA.

Principal Component Analysis: We search for the variable which has a maximum variance, which helps to find the data with the higher spread. The line on which there is a maximum variance helps to incline the data over it.

This converts the 2D data into the 1D form. This technique tries to conserve the linear structure.

19. What is a one-hot encoding?

In this technique, the representation of words is done as a group of vectors of a total number of words.

For example: The man lives in the street. The arrangement of words is done in an ascending order such as in lives man street the. Now the order in which word comes in a sentence is marked 1 else 0.

Total number of words: 5 hence vector size is 5. In the end, the vectors of each word are joined to give a representation of a whole sentence.

In lives  man  street the
The 0 0 0 0 1
Man 0 0 1 0 0
Lives  0 1 0 0 0
In  1 0 0 0 0
Street 0 0 0 1 0

This method does not work well with the large dataset and proves ineffective.

20. Name different methods of using single, multiple GPUs.

GPU can be used in the following ways:

  • Device Placement Logging
  • Manual Device Placement
  • Optimizing TensorFlow GPU Memory
  • Using a single GPU on a multi-GPU system
  • Using multiple GPUs

21. What is the hashing trick in TensorFlow?

In the process of word embedding, the text is converted to a vector using the hashing trick. The hash function assigns words to a hashing space having an index. The hash function takes five parameters named as text, n, hash_function, filter, lower, and split.

22. Name the types of neural networks.

There are two types of Neural Networks namely:

  1. Artificial Neural Network
  2. Biological Neural Network

23. What are artificial neural networks?

The artificial neural network is a representation of neurons in the brain as a nodal unit. This network helps to simplify the complex issues a user faces. It reduces the problems to a user-level to solve it.

It helps to identify objects as a human does, with its algorithmic capabilities and then predict the features through labeling. We apply it to real-time operations where data is fed in any format possible.

24. Explain cross-entropy.

Cross entropy measures the performance of an event for which the probability p and q are defined for the sample of time.

25. Provide the code for manual placement of GPU.

with tf.device('/cpu:0'):  
x = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='x')  
y = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='y')  
z = tf.matmul(x, y)  
sess = tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(log_device_placement=True))   
print(sess.run(z))

26. What is the learning rate?

The step size in the iterations which is used to determine the minimum loss is the learning rate. It decides the value after weighing the convergence and overshooting. If the value is too low, it proceeds very slowly, and if it is high, it shows sudden diverging behavior.

27. What is a dead unit?

The neuron which does not deflects from its value during the training. This situation arises when a negative value always shows 0 after passing through the ReLU layer.

28. Name the models we use in words to vector conversion.

The models which perform word embedding are:

  • Bag of words
  • Skip-gram model

29. What is bias-variance decomposition?

Bias is the error that arises due to assumptions that lie far away from the real model.
Variance is the error that arises due to fluctuations in the training set.
Bias-variance decomposition is the tradeoff between the given errors to build a generalized model with minimum errors.

30. Explain the concept of logistic regression with TensorFlow.

Logistic regression is an algorithm for the categorization of data into discrete classes. The algorithm brings the non-linearity in the data with the help of a sigmoid function. It follows binary values for the classification of data.

In TensorFlow, we define the values and pass them through the sigmoid layer activation function and train it.

Conclusion

The questions in the TensorFlow Interview Questions and Answers article are the most important and essential questions rotating in the interview rounds. This will help you to check your performance in the field of TensorFlow.

Interview questions is the starting point to expand your knowledge and dig up the place in the selection processes. These interview questions opened a room to analyze your potential and channel it in an appropriate manner.

Hope you liked the article. Do not forget to share TechVidvan rating at Google.