Neighborhood AI – where social science meets machine learning and artificial intelligence

The goal of the Neighborhood AI project is to explore the use of machine learning and general artificial intelligence in neighborhood research by combining computer vision with neural networks. The aim is to better understand the connection between the perceived environments and the social life within neighborhoods.

The project is lead by Rolf Lyneborg Lund and solely run with the passion and interest of students and researchers alike.

One of the first tasks in the project has been to automate image scraping via the Google Street View API and identifying images of houses. To reduce the burden of the model, it is imperative to reduce the image load by sorting out images that has no houses on them. The overall ratio seems to be close to 25% images of houses to 75% with no houses. It is relatively easy to define cases where there definitely are houses and where there is none. The problem arises when the cases become more intricate, e.g. where a house is maybe hiding behind the shrubbery or where something that once were a house now function as a barn. Even the simpler question “What is a house?” becomes harder to answer when wanting to automate.

All code will be available for the public as the project evolves.

Contact information

Assistant professor Rolf Lyneborg Lund: Phone: +45 9940 2846; E-mail: rolfll@socsci.aau.dk

Facts

Start date: 1 September 2018

Go to Neighborhood AI's website