Building an autoencoder

Autoencoders have several interesting applications like anomaly detection or image denoising. They aim at producing an output identical to its inputs. The input will be compressed into a lower dimensional space, encoded. The model then learns to decode it back to its original form.

You will encode and decode the MNIST dataset of handwritten digits, the hidden layer will encode a 32-dimensional representation of the image, which originally consists of 784 pixels (28 x 28). The autoencoder will essentially learn to turn the 784 pixels original image into a compressed 32 pixels image and learn how to use that encoded representation to bring back the original 784 pixels image.

The Sequential model and Dense layers are ready for you to use.

Let's build an autoencoder!

This exercise is part of the course

Introduction to Deep Learning with Keras

Exercise instructions

Create a Sequential model.
Add a dense layer with as many neurons as the encoded image dimensions and input_shape the number of pixels in the original image.
Add a final layer with as many neurons as pixels in the input image.
Compile your autoencoder using adadelta as an optimizer and binary_crossentropy loss, then summarise it.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Start with a sequential model
autoencoder = ____

# Add a dense layer with input the original image pixels and neurons the encoded representation
autoencoder.add(____(____, input_shape=(____, ), activation="relu"))

# Add an output layer with as many neurons as the orginal image pixels
autoencoder.add(____(____, activation = "sigmoid"))

# Compile your model with adadelta
autoencoder.compile(optimizer = ____, loss = ____)

# Summarize your model structure
____

Edit and Run Code

This exercise is part of the course

Introduction to Deep Learning with Keras

IntermediateSkill Level

4.8+

Start Course for Free

In this first chapter, you will get introduced to neural networks, understand what kind of problems they can solve, and when to use them. You will also build several networks and save the earth by training a regression model that approximates the orbit of a meteor that is approaching us!

Exercise 1: What is Keras?Exercise 2: Describing Keras Exercise 3: Would you use deep learning?Exercise 4: Your first neural network Exercise 5: Hello nets!Exercise 6: Counting parameters Exercise 7: Build as shown!Exercise 8: Surviving a meteor strike Exercise 9: Specifying a model Exercise 10: Training Exercise 11: Predicting the orbit!

By the end of this chapter, you will know how to solve binary, multi-class, and multi-label problems with neural networks. All of this by solving problems like detecting fake dollar bills, deciding who threw which dart at a board, and building an intelligent system to water your farm. You will also be able to plot model training metrics and to stop training and save your models when they no longer improve.

Exercise 1: Binary classification Exercise 2: Exploring dollar bills Exercise 3: A binary classification model Exercise 4: Is this dollar bill fake ?Exercise 5: Multi-class classification Exercise 6: A multi-class model Exercise 7: Prepare your dataset Exercise 8: Training on dart throwers Exercise 9: Softmax predictions Exercise 10: Multi-label classification Exercise 11: An irrigation machine Exercise 12: Training with multiple labels Exercise 13: Keras callbacks Exercise 14: The history callback Exercise 15: Early stopping your model Exercise 16: A combination of callbacks

In the previous chapters, you've trained a lot of models! You will now learn how to interpret learning curves to understand your models as they train. You will also visualize the effects of activation functions, batch-sizes, and batch-normalization. Finally, you will learn how to perform automatic hyperparameter optimization to your Keras models using sklearn.

Exercise 1: Learning curves Exercise 2: Learning the digits Exercise 3: Is the model overfitting?Exercise 4: Do we need more data?Exercise 5: Activation functions Exercise 6: Different activation functions Exercise 7: Comparing activation functions Exercise 8: Comparing activation functions II Exercise 9: Batch size and batch normalization Exercise 10: Changing batch sizes Exercise 11: Batch normalizing a familiar model Exercise 12: Batch normalization effects Exercise 13: Hyperparameter tuning Exercise 14: Preparing a model for tuning Exercise 15: Tuning the model parameters Exercise 16: Training with cross-validation

It's time to get introduced to more advanced architectures! You will create an autoencoder to reconstruct noisy images, visualize convolutional neural network activations, use deep pre-trained models to classify images and learn more about recurrent neural networks and working with text as you build a network that predicts the next word in a sentence.

Exercise 1: Tensors, layers, and autoencoders Exercise 2: It's a flow of tensors Exercise 3: Neural separation Exercise 4: Building an autoencoder

Current Exercise

Exercise 5: De-noising like an autoencoder Exercise 6: Intro to CNNs Exercise 7: Building a CNN model Exercise 8: Looking at convolutions Exercise 9: Preparing your input image Exercise 10: Using a real world model Exercise 11: Intro to LSTMs Exercise 12: Text prediction with LSTMs Exercise 13: Build your LSTM model Exercise 14: Decode your predictions Exercise 15: Test your model!Exercise 16: You're done!