Together We Stand

[ 95 ] such QCRI removes personal identifiable data on our outputs. Machine intelligence is used to learn how to automatically cate- gorize different classes of messages. All of these decisions are analysed and curated for ‘insights’. The microtask decisions are then used to train AIDR to learn from the human computing. This training improves the software’s machine learning capa- bilities. MicroMappers functions with a number of ‘clickers’ based on the type of data. There are text, image, geolocation, video, aerial and air video clickers. MicroMapper Clickers and AIDR have multiple data sources. Each of these data sources are collated to create information products. Image and text data was also incorporated from GDELT and Hemant Purohit. 6 To inform our work, we have also conducted tests with the World Bank (aerial imagery), the Global Database of Events, Language, and Tone (media information), Translators without Borders (real-time translation), Unicef (SMS data), EPFL and Drone Adventure (aerial imagery for environmental conserva- tion) 7 and UNESCO (social data). We continue to research ways to add layers of data from various social and imagery sources. Each of these projects provides us with insights to improve and extend algorithms that could one day help with overall insights or even actionable data. Our current research focuses on curated datasets and computer vision algorithms using aerial imagery. The use of unmanned aerial vehicles (UAVs) plays a pivotal role in humanitarian efforts. UAVs provide humanitarians with a bird’s eye view of the disaster-prone areas which need immediate help. These unique images can be used to determine the overall destruction that has occurred in the distressed areas as well as the most effective allocation of limited relief resources by humani- tarian organizations. Both the United States Federal Emergency Management Agency and the European Commission’s Joint Research Centre have noted that aerial imagery will play an important role in disaster response and present a big data chal- lenge. The World Bank, for example, took the initiative to cooperate with the humanitarian UAV network, UAViators, in the wake of cyclone Pam, a category 5 cyclone that razed the islands of Vanuatu in March 2015, and deployed MicroMappers. To get ahead of this big data challenge, we use the hybrid crowdsourcing and machine learning solution to rapidly process large volumes of aerial data for disaster response in a time-sensi- tive manner. 8 Humanitarian organizations make a call for digital volunteers to annotate features of interest in aerial images. For instance, in the case of typhoon Haiyan, volunteers traced healthy and damaged coconut trees whereas, in the case of cyclone Pam, volunteers assessed the level of damage for each individual build- ing in images. These human-annotated features are then used to train supervised machine learning models to recognize such features in new, unseen aerial images automatically, reducing the need for further human annotation for massive amounts of aerial image data. Such a hybrid solution for aerial image analysis has applications beyond disaster response, such as wildlife protec- tion, human rights and archaeological exploration. This research partnership with AIDR and MicroMappers garnered a few important findings. On the technology front, we have created algorithms for hybrid intelligence/computation (synergistic human+machine collaboration) based analysis of text/images/imagery/video. In addition, we have created self- learning models for damage assessment of insured structures (roofs, buildings and so on) from historical data. The tools and the practices around the data can be applied to non-crisis situ- ations. For example, we are considering how these can apply to disaster risk mitigation in urban informatics or even environ- mental sensing. Social data can be citizen engagement when applied with context. As the World Humanitarian Summit and Sustainable Development Goals indicate, there is an increased need for more local engagement and new communication processes. By having strong domain expertise and volunteer surge support, we were able to stress-test the tools and refine analysis to have more impact. Partners innovate best when they coordinate information product sharing and continuously improve on the original design. Exploring how social data can be used in partnership with official and volunteer organizations requires building a common language. Over the past three years, we have refined the type of information the tools collect. Data is not useful without anal- ysis. It is important to incorporate domain knowledge and a holistic view of the situation using digital forensics to achieve better results. One big opportunity is that the greater digital community (local and global) will join in efforts to iterate on new science to support outcomes to benefit the humanitarians and, in time, the affected populations. AIDR and MicroMappers analyse various types of data including social media messages, SMS, imagery, text and images Tagger Collector T ogether W e S tand