In this chapter, we learned about algorithmic pareidolia in computer vision. We have explained how CNN models can be interpreted by various visualization techniques, like forward pass-based activation visualization, gradient ascent-based filter visualization. Finally, we introduced the DeepDream algorithm, which again, is a slight modification of the gradient ascent-based visualization technique. The DeepDream algorithm is an example of transfer learning being applied to a computer vision or an image-processing task.
We will see more similar applications in the next chapter, which will be focusing on style transfer.
