Can we train deep learning models to identify distressed housing using street-level images? Yes: I developed a model that could predict, with 87% accuracy, whether a house was on its way to demolition. More details here.
As an initial experiment to show ConvNets using civic data and Google Street View images can help predict urban health risks, I trained a model to identify roof types in Boston, MA. It could distinguish between flat and gabled roofs with 92% accuracy. I explain the experiment here.
A serious threat to low-income housing in U.S. cities, especially in hot rental markets like Boston, is the expiration of the contracts that enable Section 8 subsidized units. I built the visualization below using data from preservationdatabase.org. Navigate to find out when and where the greatest prospective loss exists, for Boston and nearby Cambridge and Somerville.
Is the apparent ‘Trump Effect’ even bigger than news outlets reported?
The FBI’s annual report on hate crimes in the U.S. showed anti-Muslim hate crimes jumped sharply in 2015: over a one-year span, reported incidents rose from 154 to 257.
This news bolsters the case for a 'Trump Effect': hostility stirred up and/or channeled by the President-Elect against Muslims, Latinos, immigrants, women, transgender folks, and others. It’s consistent with data...
That makes three. The first, in Miami’s Wynwood district, ended in September after 45 days with no confirmed transmission. The second, in Miami Beach, has remained active since August. (With frustration mounting, its city commission has asked permission to...