The document discusses how deep learning could automate cities by collecting various types of data from IoT sensors and using machine learning models. It provides examples of data that could be collected, describes the machine learning pipeline of data collection, feature extraction, and model learning. Potential use cases for deep learning in cities include predicting crime, optimizing transportation, analyzing franchise locations, predicting smog, alleviating traffic, better city planning, and better political policies.
11. The brain’s visual system has 10.e14neural connections. And you only live for 10.
e9seconds. So it’s no use in learning one bit per second. You need more like 10.
e5bits per second. And there’s only one place you can get that much information:
from the input itself.
- Geoffrey Hinton
20. What did they actually do?
- gather data from open data sites of Chicago, San Francisco and some additional
data about weather and census
- preprocess data
- build one big matrix where rows correspond to crimes examples and columns to
some features like location, type of crime, area, date etc.
- split data to training, validation and test sets
- train Deep Neural Network to predict probability of arrest for a given crime