“Let the computer dream while it sleeps”
One of the most interesting paradigms that differentiate humans from machines are dreams! Of course, everyone right out there dreams, either in his sleep or when he is still awake (the latter part suits me well :p ). Some months ago, there was a news saying that a machine passed the Turing Test! And now, machines are on their way to prove themselves as independent creatures. Thanks to Google Research Team for unveiling the fact that machines can indeed dream .It’s time to present you the facts technically .
Artificial Neural Networks have spurred remarkable recent progress in image classification and speech recognition. We train an artificial neural network by showing it millions of training examples and gradually adjusting the network parameters until it gives the classifications we want. The network typically consists of 10-30 stacked layers of artificial neurons. Each image is fed into the input layer, which then talks to the next layer, until eventually the “output” layer is reached. The network’s “answer” comes from this final output layer. Google Research Team tried to visualize the results of the image after it passed each hidden layer, which as a result made us see the computer dream.
At each level, the neural net interprets the image differently based on its training module and we ask more of it. As a result, it shows us new visuals, which emerge out of a single picture. This is actually useful in detecting where the neural nets are making a mistake. For example, there were images that the neural nets generated for a dumbbells, which had an arm in every picture. This again increases the ambiguity of the system, which understood that dumbbells and human arms are inseparable. This is again wrong , as the system understood that the two can never exist separately!
Two ways to see your machine dream :
1) If you’re really a geek and love experimenting stuffs, then here is a good news. Google released their deep-learning code open-source in their Github repository. Check out from https://github.com/google/deepdream
2) If you are in short of time, space and energy ,then you should probably check out http://psychic-vr-lab.com/deepdream/. Upload your pic and wait until the computer dreams!