Google’s neural network to compose email responses for you
This article discusses how Google is using deep neural networks to propose automatic replies based on the content of the email. The technology will analyze emails and produce three most sensible responses to the content of the email. This will allow mobile users to quickly respond to emails rather than wasting time typing on a touchscreen.
The technology used is called sequence-to-sequence learning where two neural networks fuse understanding a language and synthesizing language. The resulting network uses vector transformation to understand the email content. A separate network then comes up with responses based on the first network. One of the main challenges was to ensure that the responses are not similar to each other. Another major challenge was decoding the various styles and tones people use to describe the same thing.
This relates to networks because it deals with the network effects that researchers have to deal with when decoding actual content with computers. This type of network uses some sort of ranking procedure to determine the second network of responses to see which are the three best responses that the user would most likely choose. This is extremely relevant to course material because the algorithm analyzes the network structure and determines another network based on the first network. This procedure is kind of like PageRank and Hits algorithms where the results are based on rankings based on the sources.
http://www.gizmag.com/google-neural-network-ai-response-mail/40201/