Search for HOT projects, news, people and jobs.
News — 20 March, 2019

Bringing machine learning and open data to the Tasking Manager

Addressing the unmapped gap with higher quality and greater efficiency. Photo above from the Tasking Manager meeting held last month in Washington, DC.

Last month we kicked off a new phase for the Tasking Manager – building a community of collaborators for software development. The community is addressing usability and user flows, metrics and statistics, and machine learning. Today we’re sharing about a new partnership with Microsoft Philanthropies and Bing Maps to bring design updates, machine learning integration, and new open building datasets into the Tasking Manager through Microsoft’s AI for Humanitarian Action program.

Why does this matter?

Hundreds of millions of map features critical for humanitarian responses are missing from OpenStreetMap. Since the inception of the Tasking Manager, the HOT community has mapped at an incredible rate — 11 million square kilometers mapped in Africa alone. However, large parts of Africa, a total land area of 30 million square kilometers, with vulnerable populations to disasters 60% still remains unmapped. As we look to address this gap, the process of how we map OpenStreetMap through the Tasking Manager needs to be enhanced.

At the HOT Summit in September 2018, we held a workshop about machine learning integration into a Tasking Manager mapping workflow. A key point of the discussions at the workshop was the appropriate use of machine learning outputs into the mapping workflow. To meet the challenge of cover more parts of the world, we identified the Tasking Manager requires two key improvements in the mapping workflow: better data quality of edits and greater efficiency of mapping and validation.

The partnership with Microsoft comes out of the discussions at the HOT Summit and later work at SatSummit to help bring developer and design resources to OSM and enable machine learning workflows. In addition to the machine learning updates to the Tasking Manager, Microsoft will be creating country-wide open building datasets by building upon their work in the US and Canada. This will be one of the first open building datasets in Africa that will be available for use within OSM.

What you’re going to see worked on over the life of the project?

Over the next seven months, the partnership will work on the following:

  1. Conduct design research for improved mapping workflows and enhance Tasking Manager with machine learning

  2. Build a machine learning API for integrating outputs of machine learning models into the Tasking Manager

  3. Create country-wide open building datasets in Uganda and Tanzania

  4. Pilot a feedback loop to help improve machine learning models and outputs

In February, we kicked off the first two items and you’ll begin to see outputs over the next month. This work is a part of the increase in development on the Tasking Manager from contributing organizations and we’re having bi-weekly working group meetings and active Slack discussions – see the #ml4tm and #tasking_manager_3 channels.

What’s Next?

A couple near term items to keep on the lookout:

  1. User flow research from the design group Major. We’ve been working with a Lisbon-based interaction design firm called Major to research and evaluate how users move through the system.

  2. A deep dive into Tasking Manager data. As we’ve been working with Major, we’ve been looking into the data about task completion, validation, and other metrics about speed and coverage of mapping.

All this new work is thanks in part to the American Red Cross and the work they’ve done with Microsoft through the Missing Maps project. American Red Cross is a key partner in the project and will be helping work with us as we implement the work on the Tasking Manager.

Keep watching this blog for more updates on the project and other Tasking Manager work!