DLT Unit-03

💡 Click on any image for better visualizing!

1) Neural Networks

A neural Network is a type of Machine Learning Model inspired by the human brain
Deep Learning uses a neural network for learning the model
There are so many types of neural networks. Some of them are:
- a)ANN
- b)CNN
- c)RNN

ANN	CNN	RNN
ANN stands for Artificial Neural Network	CNN stands for Convolutional Neural Network	RNN stands for Recurrent Neural Network
ANN is also called a Feed-Forward Neural Network(FFNN)	It uses a Convolutional layer to extract features from the data	It uses a recurrent layer to process the sequential data
Input Layers -> Hidden Layers -> Output Layers	It processes grid-like data	It processes Sequential data
Input data is Independent	Input data is independent	Input data is dependent
It handles fixed-length inputs	It handles fixed-length inputs	It handles Variable-length inputs
ANN is very simple	CNN is less complex	RNN is highly complex
Example: Pattern Recognition	Example: Image Recognition	Example: Time-series prediction

DLT Unit-03

2) Anatomy of Neural Network

Layer:

Layer is the Modular units that process the data
Layers can be stateless or stateful
Layer types:
- Convolutional Layers
- Recurrent Layers

Model(Network):

It is a DAG of layers that maps inputs to outputs
They are built using:
- Sequential API
- Functional API
Sequential API is used for simple models
Functional API is used for complex models

Input and Target:

They must be properly shaped before feeding into the Network

Loss Functions:

Loss function is the difference between the predicted value and the actual value

Optimizer:

Optimizer updates the weight based on the loss functions

DLT Unit-03

3) Deep Learning Stack

Deep Learning Stack consists of multiple layers of software and hardware
They work together to build, train and deploy the neural network model

a) Hardware Stack:

It contains Hardware Components like:
- CPU(Central Processing Unit)
- GPU(Graphical Processing Unit)
- Memory, etc...
Example:
- NVIDIA

b) Software Stack:

Libraries:
- Libraries are used to perform tensor operations and computations
- Example:
  - TensorFlow
Frameworks:
- Frameworks are used to build the model
- Example:
  - Keras
Programming Language:
- We use Python here

DLT Unit-03

4) Architecture of Keras (or) Keras Workflow (or) high-level building blocks

Prepare the Data:

Create Input Sensor and Target Sensor
- Input Sensor: x_train
- Target Sensor: y_train

Layer:

Layer is the Modular units that process the data
Layers can be stateless or stateful
Layer types:
- Convolutional Layers
- Recurrent Layers

Model(Network):

It is a DAG of layers that maps inputs to outputs
They are built using:
- Sequential API
- Functional API
Sequential API is used for simple models
Functional API is used for complex models

Compile the Model:

model.compile(optimizer = 'Adam',
- loss = 'value',
- metrics = ['accuracy'])

Train the Model:

model.fit(x_train,y_train,
- epochs = 10,
- batch_size = 512,
- validation_data = (x_val,y_val))

Evaluate the Model:

model.evaluate(x_test,y_test)

Predict the Model:

model.predict(new_inputs)

DLT Unit-03

5) Binary classification(classifying movie reviews)

Binary classification for movie reviews involves classifying each movie review as positive or negative.

Input:

Use the IMDb dataset.
It contains 25,000 training samples and 25,000 testing samples.

Pre-Processing:

Transform all text into lowercase and remove all special characters.
Convert the sentences into a list of tokens.

Feature Extraction:

Use Keras's Embedding Layers to convert tokens into vectors.
Example:
- a = 100
- b = 10
- Embedding(a, b)

Classification:

After the Embedding Layer, add a Flatten Layer to reduce the dimensions.
Add one or more Dense Layers.
Add a final Dense Layer for binary output.
The model classifies each movie review as positive or negative.

DLT Unit-03

6) Training Loss vs Validation Loss

program:

a = history.history
b = a['loss']
c = a['val_loss']
epochs = range(1, len(b) + 1)

plt.plot(epochs, b, 'bo', label='Training loss')
plt.plot(epochs, c, 'b', label='Validation loss')

plt.title('Graph')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()

Output:

The difference between the training loss and validation loss indicates Overfitting

DLT Unit-03

7) NVIDIA GPUs vs cloud-based GPUs

NVIDIA GPUs	cloud-based GPUs
NVIDIA GPUs are the best GPUs for deep learning	When you don't have an NVIDIA GPU, Cloud-based GPUs are best for deep learning
NVIDIA GPUs support frameworks and libraries such as TensorFlow and Keras.	Cloud-based also supports frameworks and libraries such as TensorFlow and Keras.
NVIDIA GPUs support CUDA	Cloud-based GPUs also support CUDA
It needs an Operating System	It needs an Internet connection
NVIDIA GPUs are better when compared to other GPUs	Cloud-based GPUs are better than NVIDIA GPUs
High performance and High Throughput when compared to other GPUs	High performance and High Throughput when compared to NVIDIA GPUs
High Cost	cost effectiveness

DLT Unit-03

8)Explain the steps in setting up the deep-learning workstation.