wayground logo

Free Printable Worksheets

NEW

Font size

S
M
L
XL
Worksheets

Deep Learning — III B.Sc. Data Science

Total questions: 80

Worksheet time: 40mins

Name
Class
Date
1.

Who is considered the father of Artificial Intelligence?

a)

Alan Turing

b)

Andrew Ng

c)

Elon Musk

d)

Stuart Russell

2.

The term “Artificial Intelligence” was coined at:

a)

MIT Conference

b)

Dartmouth Conference

c)

Stanford Symposium

d)

Oxford AI Summit

3.

What were early rule-based AI systems called?

a)

Intelligent Agents

b)

Expert Systems

c)

Neural Networks

d)

Deep Agents

4.

Machine learning algorithms “learn” from:

a)

Rules

b)

Data

c)

Hardware

d)

Human intuition

5.

Early neural network training breakthrough came from:

a)

CNNs

b)

Backpropagation

c)

NLP

d)

GANs

6.

ML boom restarted in 2010s because of:

a)

More textbooks

b)

More powerful GPUs

c)

AGI inventions

d)

Bigger keyboards

7.

What enabled modern deep learning?

a)

Tiny datasets

b)

Cloud gaming

c)

Large datasets

d)

Analog computers

8.

Who developed Convolutional Neural Networks?

a)

LeCun

b)

Bengio

c)

Ng

d)

Turing

9.

A Deep Neural Network has:

a)

Only 1 layer

b)

No activations

c)

Multiple hidden layers

d)

Only input

10.

Node in neural network =

a)

Neuron

b)

Bit

c)

Token

d)

Window

11.

What do weights represent?

a)

Hyperparameters

b)

Learnable parameters

c)

Accuracy

d)

Inputs

12.

Activation functions introduce:

a)

Linearity

b)

Non-linearity

c)

Symmetry

d)

Memory

13.

Most common hidden layer activation:

a)

Sigmoid

b)

ReLU

c)

Softmax

d)

Constant

14.

Softmax is used in:

a)

Regression

b)

Multi-class classification

c)

Clustering

d)

GANs

15.

Loss function measures:

a)

Hardware

b)

Difference between predicted & true

c)

Dataset length

d)

GPU usage

16.

Gradient descent updates:

a)

Layers

b)

Weights

c)

Neurons

d)

Batch size

17.

Backpropagation propagates:

a)

Random numbers

b)

Output classes

c)

Gradients

d)

Training time

18.

Dropout is used for:

a)

Overfitting reduction

b)

Faster inference

c)

Memory expansion

d)

Model cloning

19.

Which network is best for image data?

a)

CNN

b)

LSTM

c)

RNN

d)

Transformer

20.

Which network uses sequence memory?

a)

CNN

b)

LSTM

c)

Logistic Regression

d)

GAN

21.

Autoencoders are used for:

a)

Classification

b)

Feature learning

c)

Reinforcement

d)

Sorting

22.

Batch size means:

a)

Total dataset

b)

Labels only

c)

Samples per training step

d)

Number of hidden layers

23.

TensorFlow was developed by:

a)

Amazon

b)

Google

c)

Meta

d)

Apple

24.

Which library is NOT primarily for ML?

a)

NumPy

b)

TensorFlow

c)

Keras

d)

PyTorch

25.

Tensor =

a)

Loss

b)

Multi-dimensional array

c)

GPU

d)

Model path

26.

TensorFlow Variables can:

a)

Never change

b)

Update during training

c)

Store only images

d)

Store only integers

27.

TensorFlow Graph stores:

a)

Image maps

b)

Computational operations

c)

Python backtraces

d)

Loss reports

28.

GradientTape is used to:

a)

Record gradients

b)

Play music

c)

Store checkpoints

d)

Test accuracy

29.

TensorFlow Dataset API helps with:

a)

Visualization

b)

Data pipeline

c)

GPU drivers

d)

Loss reduction

30.

Epoch means:

a)

1 mini-batch

b)

1 full pass over dataset

c)

1 layer

d)

Loss multiplication

31.

Optimizer changes:

a)

Dataset

b)

Weights

c)

Batch size

d)

Number of neurons

32.

Adam stands for:

a)

Adaptive Moment Estimation

b)

Automatic Model Expansion

c)

Adaptive Model Encoder

d)

Adaptive Memory

33.

Loss function is used to:

a)

Test GPUs

b)

Measure prediction error

34.

Optimizer purpose

a)

Reduce loss

b)

Generate samples

c)

Save checkpoints

d)

Create graphs

35.

Training loop steps order

a)

Loss → Predict → Update

b)

Predict → Loss → Update

c)

Train → Deploy → Save

d)

Test → Train → Save

36.

Validation is used to detect

a)

Underfitting only

b)

Overfitting

c)

GPU levels

d)

Activation failure

37.

Keras is a

a)

Low-level API

b)

High-level API

c)

Database

d)

Browser engine

38.

Layers are added using

a)

model.compile

b)

model.add

c)

model.predict

d)

keras.layer.new

39.

sequential() model type is

a)

Graph-based

b)

Linear stack of layers

c)

RNN-only

d)

GAN-only

40.

Model training in Keras uses

a)

model.fit

b)

model.build

c)

model.save

d)

model.image

41.

What does a neuron compute in a neural network?

a)

Input only

b)

Weighted sum + activation

c)

Batch size

d)

Loss function

42.

What is a layer in a neural network?

a)

A CSV file

b)

A group of neurons

c)

A dataset partition

d)

A training loop

43.

The output of a neuron before activation is called

a)

Bias

b)

Z-value

c)

Alpha

d)

Time step

44.

Bias helps the model

a)

Reduce learning rate

b)

Shift activation function

c)

Remove neurons

d)

Reduce overfitting

45.

Sigmoid outputs range from

a)

-1 to 1

b)

0 to 1

c)

0 to 10

d)

-10 to 10

46.

Which activation avoids exploding outputs?

a)

Linear

b)

ReLU

c)

Tanh

d)

HardMax

47.

Why is ReLU preferred?

a)

No training needed

b)

Fast computation and avoids vanishing gradients

c)

Safe from overfitting

d)

Built only for GPUs

48.

What does Softmax output?

a)

Binary values

b)

Probabilities

c)

Loss values

d)

Gradients

49.

Overfitting occurs when model

a)

Performs well on train and test

b)

Performs poorly on train

c)

Performs well on train but poorly on test

d)

Has many layers

50.

Early stopping prevents

a)

Underfitting

b)

Overfitting

c)

GPU crashes

d)

Model saving

51.

A shallow neural network has

a)

Many hidden layers

b)

5+ input layers

c)

Zero hidden layer

d)

2 output layers

52.

Which improves generalization?

a)

Large batch

b)

More epochs

c)

Data augmentation

d)

Higher dropout rate always

53.

TensorFlow is mainly used for

a)

SEO

b)

ML and Deep Learning

c)

Blockchain

d)

3D rendering

54.

PyTorch is popular because of

a)

Static graph

b)

Dynamic graph

c)

Low RAM

d)

Web hosting

55.

Which library uses .nn.Module?

a)

TensorFlow

b)

PyTorch

c)

Scikit

d)

Keras

56.

Scikit-learn is used mainly for

a)

Classical ML

b)

Hardware control

c)

Visualization

d)

Database indexing

57.

Which library is used heavily for numerical operations?

a)

NumPy

b)

SQLAlchemy

c)

Selenium

d)

Scikit-image

58.

TensorBoard helps with

a)

GPU drivers

b)

ML model visualization

c)
59.

PyTorch Lightning provides:

a)

High-level training wrapper

b)

Compiler

c)

Dataset generator

d)

Browser app

60.

JAX differentiates using:

a)

AutoGrad

b)

CNN

c)

Decision trees

d)

BERT

61.

TensorFlow tensors support:

a)

Only 1D

b)

2D only

c)

Multi-dimensional arrays

d)

Only boolean values

62.

In TF, tf.constant is:

a)

Mutable

b)

Unchangeable

c)

A layer

d)

A model

63.

In TF, tf.Variable is:

a)

Static

b)

Read-only

c)

Trainable

d)

GPU-only

64.

The computational graph represents:

a)

Folder structure

b)

Flow of mathematical ops

c)

GPU drivers

d)

File permissions

65.

Eager execution is:

a)

On-demand execution

b)

Lazy loading

c)

Graph only

d)

GPU emulation

66.

tf.cast() is used to:

a)

Change tensor type

b)

Load datasets

c)

Save checkpoints

d)

Train model

67.

tf.one_hot() converts labels into:

a)

Images

b)

Text

c)

Binary vectors

d)

Gradients

68.

tf.argmax() returns:

a)

Minimum element

b)

Index of max element

c)

Loss

d)

Number of layers

69.

tf.metrics contains:

a)

Activation functions

b)

Evaluation metrics

c)

GPU tools

d)

Datasets only

70.

Model building step order:

a)

Define → Compile → Train

b)

Compile → Train → Define

c)

Evaluate → Save → Train

d)

Train → Deploy → Define

71.

Why compile a model?

a)

Allocate memory

b)

Select loss and optimizer

c)

Convert data

d)

Create graph

72.

In supervised learning, labels are:

a)

Random samples

b)

Unknown outputs

c)

Ground truth

d)

Hidden layers

73.

Validation helps identify:

a)

GPU faults

b)

Overfitting

c)

Model import

d)

Input loss

74.

Batch normalization helps:

a)

Reduce data size

b)

Stabilize training

c)

Remove neurons

d)

Increase epochs

75.

Learning rate controls:

a)

#layers

b)

Step size in weight updates

c)

Tensor shape

d)

Batch size only

76.

Training stops when:

a)

Loss is 0 only

b)

Early stopping triggers

c)

GPU overheats

d)

Labels change

77.

Keras API type:

a)

Low-level

b)

High-level

c)

Hardware API

d)

Cloud API

78.

Layers in Sequential model are added via:

a)

model.layer()

b)

model.add()

c)

model.connect()

d)

keras.link()

79.

Keras Functional API is best for:

a)

Simple stacks

b)

Complex multi-input/multi-output models

c)

Only RNN

d)

YOLO architectures only

80.

model.save() stores:

a)

Dataset

b)

Entire model architecture + weights

c)

GPU config

d)

Compilation logs