Imagine that your local ambulance and fire dispatch runs on an algorithm developed with the help of the data scientists who built Lyft’s grid-optimization system.
Former Eventbrite lead data scientist Paul Duan, nonprofit veteran Andrew Jiang, and former Thomvest Ventures analyst Eric Liu founded Bayes Impact earlier this year.
Bayes Impact’s mentors and advisors hail from Airbnb, Ayasdi, Intuit, LinkedIn, Netflix, OpenTable, Salesforce.com, and Square. Its advisors include Lyft chief technology officer Chris Lambert, Airbnb head of data science Riley Newman, and Intuit vice president of data science Chris Chapo. “On the mentor side, it’s a much longer list; we just haven’t put everyone’s up yet,” Jiang said in an interview with VentureBeat.
More and more non-profits are using data science to solve social problems.
DataSift is partnering with United Nations Global Pulse to analyze social data for humanitarian purpose. Code for America is working on reducing emergency-room calls in Long Beach, Calif., using big data. Rayid Ghani, formerly chief data scientist on the Obama for America campaign, has been running the Data Science for Social Good fellowship for two years.
“Almost all technology companies, no matter their size, use data to their users’ advantage: not-for-profits and governments can and should do the same thing,” Zachary Townsend, founder of the startup Standard Treasury and a Bayes Impact board member, wrote in an email to VentureBeat.
The fellowship takes six to 12 months. It provides San Francisco housing and a stipend.
“We are still thinking about what the exact dollar amount would be, [but] we are thinking it would likely be somewhere between $4,000 to $6,000 per month as a living stipend, kind of based on experience,” Jiang said.
The fall cohort will consist of 20 to 25 fellows working on 10 to 12 projects. Fellows are expected to come with enough experience to start working immediately.
“The fellows are generally either post doc, Ph.D. students in, like, physics, math, computer science, or they are experienced data scientists themselves,” said Jiang.
Candidates for the fall program include a quantitative trader with a master of science degree in math from Oxford, a postdoctoral researcher at the Max Planck Institute, and a former senior software engineer at Google.
Each Bayes Impact project will involve approximately two full-time data science fellows, with in-person mentorship coming from two or three data scientists. If necessary, Bayes Impact will also bring in content experts with specific knowledge of subjects like bioinformatics to help teams understand their projects better.
Six fellows are working on four projects in Bayes Impact’s summer pilot program. Two of them were from Zipfian Academy, participating in the program via a partnership between the two organizations.
Initiatives range from detecting fraud and assessing credit worthiness for microlending nonprofit Zidisha to cutting down on recidivism and overcrowding inside California prisons.
William Lane, who just finished a master degree in statistics at Stanford, is working with another fellow on the Zidisha project. The team’s mentors are John McDonnell and Christopher Moody — two data scientists working on fraud detection at Square.
When Lane started, the scope of the project and the amount of data he had to work with were daunting.
His mentors pointed out that it was a good idea to capture some low-hanging fruit. So the first thing he should do was just “a simple model that has a very clear goal,” Lane told VentureBeat. In the micro-lending context, for example, the idea was to look at first-time borrowers and try to eliminate those people who pay back zero.
“[The mentors at Bayes Impact are] pretty carefully selected, for the particular problems, not just somebody who has been working as a data scientist, but really knows about what problems in particular the fellow might be working on,” Lane said.
Bayes Impact has lined up some projects for the fall program. One project partner, a fire department, will work with the fellowship to predict which buildings will have the highest chance of catching fire, based on building-inspection data.
Much of the funding for Bayes Impact was bootstrapped. Organizers also secured money from a grant. Project partners pay a small amount of money to cover some expenses, too. The good part is that after hearing about Bayes Impact, a lot of funders got interested in funding it.
“Like many not-for-profits, I think that fundraising is the critical challenge,” Townsend said. “Foundations aren’t like venture capitalists, and most donors aren’t like angel investors: They want to see a thoroughly proven model before they give money.
“Bayes Impact is on the path to having very successful projects and significant impact, so I think they’ll overcome this problem.”