Scribd
Upload
21 views

Auto Encoders

This document provides an overview of autoencoders and their components. It discusses how autoencoders use an encoder to compress high-dimensional input into a lower-dimensional latent space, and then use a decoder to reconstruct the output from this latent space. The encoder learns to extract key features from the input, while the decoder aims to reconstruct the original input from these features. Different types of autoencoders are also introduced, including denoising autoencoders which are designed to clean noisy data and extract robust features.

Uploaded by

mobeen
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
21 views

Auto Encoders

This document provides an overview of autoencoders and their components. It discusses how autoencoders use an encoder to compress high-dimensional input into a lower-dimensional latent space, and then use a decoder to reconstruct the output from this latent space. The encoder learns to extract key features from the input, while the decoder aims to reconstruct the original input from these features. Different types of autoencoders are also introduced, including denoising autoencoders which are designed to clean noisy data and extract robust features.

Uploaded by

mobeen
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 23

Agenda

● Auto Encoders
○ Introduction
○ Architecture
○ Encoder
○ Feature space
○ Decoder
● Types of Decoders
● Applications of Decoders

1
Previously
● Artificial Neural Networks (ANN)
● Convolutional Neural Networks (CNN)

2
Encoders & Decoders in Deep Learning
Encoder
● Converts original data into a secret code.
● Uses rules or transformations to hide the message.

Decoder
● Reverts the secret code back to the original data.
● Understands the rules to reverse the encoding.
● Enables understanding of the hidden message.

3
What Are Autoencoders?
A type of neural network designed for dimensionality reduction and feature
learning.

Primary goal: To encode data into a compact representation and then decode it
for reconstruction.

4
Anatomy of an Autoencoder
Components: Encoder, Bottleneck/Hidden Layer, Decoder.

5
Anatomy of an Autoencoder
Components: Encoder, Bottleneck/Hidden Layer, Decoder.

6
Encoder
The encoder is like a detective that learns to capture the most important features
of an image or data.
It's the first part of the autoencoder and responsible for compressing the input data.

7
How Does the Encoder Work?
Imagine an encoder as a funnel that squeezes a big picture into a smaller
representation.
It transforms the input data (e.g., an image) into a compact form called the latent
space or encoding.

8
Encoding Process
Step 1: The input data, such as an image, is fed into the encoder.
Step 2: The encoder consists of layers of neurons that learn patterns and features
in the data.
Step 3: These patterns are combined and transformed into a compact
representation in the latent space.

9
Why is it Important?
Data Compression: The encoding is much smaller than the original data, making it
efficient to store and transmit.
Feature Extraction: The encoder learns to extract valuable information, which can
be used for various tasks.

10
What is a Decoder?
The decoder is like an artist that takes the compact representation (encoding)
from the encoder and recreates the original data.
It's the second part of the autoencoder and responsible for generating output from
the encoding.

11
How Does the Decoder Work?
Imagine a decoder as a reverse funnel that expands the compact representation
back into a full image or data.
It transforms the encoding from the latent space back into a reconstructed
output.

12
Decoding Process
Step 1: The encoded data (latent space representation) is fed into the decoder.
Step 2: The decoder consists of layers that learn to reverse the compression
process by generating features and patterns.
Step 3: These generated features are combined to reconstruct the original data.

13
Recreating Original Data
The decoder's goal is to recreate data as close to the original input as possible.
It uses the knowledge it gained from the encoder to generate meaningful features.

14
Why is it Important?
Data Reconstruction: The decoder's job is to bring back the data's original form
from the compressed encoding.
Completing the Loop: Autoencoders aim to minimize the difference between the
input and the reconstructed output.

15
Simple/Vanilla Auto-encoder

16
Types of Auto-Encoders
1. Vanilla Autoencoder
2. Denoising Autoencoder
3. Variational Autoencoder (VAE):

17
Denoising Autoencoder

18
Input Image

19
Output Image

20
Denoising Autoencoder
Denoising Autoencoders are a type of neural network.
Designed to clean noisy or corrupted data and extract essential features.

21
Denoising Autoencoder

22
Denoising Autoencoder

23

You might also like