Convolutional autoencoder

As explained in Evolutionary Deep Learning: Genetic algorithms and neural networks

A convolutional autoencoder is an autoencoder whose encoder and decoder use convolutional operations to learn and reconstruct structured inputs such as images.

A convolutional autoencoder learns to reconstruct structured inputs, especially images, without labels. It combines convolutional feature extraction with an encoder–decoder design: the encoder compresses an image into a narrow latent representation, and the decoder rebuilds the image from that representation. section 8.0 section 8.1.1

Training repeatedly compares each input with its reconstruction, commonly using mean squared error or pixel-wise loss. Convolutional and MaxPool layers progressively extract and aggregate image features; UpSampling layers expand the latent view back toward the original dimensions. section 8.1.1 section 8.1.2

A compact Keras architecture follows this sequence:

• Accept a 28×28 single-channel image. section 8.1.2
• Encode it with convolutional layers using 64, 32, and 16 filters, placing 2×2 MaxPool layers after each convolution. section 8.1.2
• Decode the latent view with convolutional layers and three 2×2 UpSampling operations. section 8.1.2
• End with a one-channel convolution, combine the input and output into a Keras model, and compile it with Adam and MSE. section 8.1.2

Visualizing learned representations

After training on handwritten digits, project the latent encodings into two dimensions with t-SNE. Clustering by digit class makes the feature representation visible even though the autoencoder received no labels. section 8.1.1

Figure 8.1 A convolutional AE

Use the architecture figure to inspect the encoder, latent view, decoder, convolutional layers, pooling, and UpSampling flow. section 8.1.1

How Evolutionary Deep Learning: Genetic algorithms and neural networks covers it

How deep it goes

Descends from architecture and latent-space intuition to a concrete Keras layer configuration, compilation, and visualization.

Examples

Uses MNIST latent-space visualization and a Fashion-MNIST Keras notebook with encoder, decoder, model summary, and training outputs.

What it assumes

Teaches autoencoder principles while extending prior familiarity with convolutional neural network architecture.

What you can do after

Builds and compiles a convolutional autoencoder that reconstructs single-channel images and exposes learned latent encodings.

How it connects

Connects autoencoders with convolutional neural networks, unsupervised learning, representation learning, latent space, and variational autoencoders.

liveBookmanning.com

Read Evolutionary Deep Learning: Genetic algorithms and neural networks on liveBook Get the book at manning.com

Related concepts

• Kerasused by · Keras is an open-source, high-level deep-learning API for building, training, evaluating, and deploying neural-network models.
• convolutional layeruses · A convolutional layer is a neural-network layer that applies shared learned filters to local regions of structured input to produce feature representations.
• autoencoderkind of · An autoencoder is a neural network that encodes input data into an internal representation and decodes it to reconstruct the input or an approximation.
• transposed convolutionuses · A transposed convolution is a learnable convolution-like operation corresponding to the transpose of a convolution operator and often used to increase spatial resolution.
• poolinguses · Pooling is an operation that aggregates values or representations across regions or axes to produce a reduced or summarized representation.
• convolutional neural networkuses · A convolutional neural network is a neural-network architecture that applies learned convolutional operations to extract patterns from structured data.

This explainer was generated by Manning's knowledge graph from the text of Evolutionary Deep Learning: Genetic algorithms and neural networks. The section links point to the parts of the book it draws on.