Machine Learning in Elixir

Using CNNs for Single-Step Prediction

Recall from Chapter 7, ​Learn to See​, that the convolutional neural networks take advantage of locality and weight sharing to effectively learn patterns on image data. Images naturally have local structure—features that are close together spatially in an image are often related. Additionally, convolutional neural networks apply the same set of parameters across an entire sliding window. That means convolutional neural networks learn to extract the same set of features from multiple regions in the input.

The properties that make convolutional neural networks useful for image data also make them useful for time-series data. With time-series data, local relationships matter because adjacent timesteps are likely ...