Advances in AI and ML have impacted our daily and work lives and at Microsoft we see unique applications of how these can be used to improve products and services. Through our work we are finding that ML can have an equally transformative effect on societally important problems such as healthcare, financial inclusion and governance. This talk discusses the potentials and challenges of using AI for societal impact projects and presents two case studies—helping health workers ensure adherence to tuberculosis medication and on enabling better online support for mental health.
Most languages in the world lack linguistic resources to build large, data hungry models and systems. To be able to truly support speech and language systems that can enable everyone on the planet, methodologies and techniques to build systems in resource constrained settings are essential. Microsoft has the technologies, expertise and reach to make this possible. Ellora aims to pull all this together into a concerted effort to tackle this challenge.
Most efforts around the world to create technology for the vision impaired focus on functional aspects of everyday living: assistance for navigation, skills for employability, unassisted shopping, independent indoor mobility etc. This talk will discuss two projects where the focus is on empowering the vision impaired to accomplish more. The first project is CodeTalk which enhances productivity of blind programmers in using IDEs like Visual Studio and CodeTal, which identifies four classes of barriers for access to IDEs namely, discoverability, alertability, navigability and glanceability, and provides solutions. The second project Video Games makes mainstream video games accessible to the blind. This is akin to the enhanced access to mainstream movies provided by audio descriptions. The talk will describe how a spatial audio based interaction toolkit could enable a blind person to play and enjoy mainstream video games.
Amphibie is an all-terrain robot that can be sent to any location during a natural disaster to help in performing rescue operations. Amphibie is capable of delivering relief materials to the victims. The bot can perform search operations in manual as well as automated mode. It can work irrespective of the network connectivity as it can work on all platforms [IoT/Bluetooth/RF]. It calculates all the parameters which are required to perform a rescue operation in an effective manner. Amphibie 2.0 can predict natural disaster with enhanced ML algorithms.
Disasters often strike when we are least prepared to face them. They leave behind a trail of destruction adversely affecting human life and property. The loss caused by disasters can be significantly reduced with better communication and proper management. Keeping this in mind we designed RVSAFE, a one stop solution for effectively handling any kind of disasters.
En-route to providing a comprehensive way to predict and manage a natural disaster to occur in future, our android application Epiphany has been designed and built using modern solutions under extensive research study. Our AI predictability accommodates the use of machine learning, neural networks and advanced NLP techniques to predict a calamity to occur in future. It also includes an offline services section containing Offline Internet and Wifi Mesh to cater to the "no internet" scenario in case a calamity strikes. Further post calamity management services like volunteering to share shelters, donation of commodities, blood bank, AR Navigation and more is also included to make this a full fledged application to be used in case of a natural calamity.
Rahat is an elegant solution that can be used by both victim and rescuer with its goal being to bridge the existing gap between the two. Rahat contacts the victim first, getting information like how their location, number of victims, etc. We also provide a Dashboard to rescue services enabling them to allocate resources better. Our platform can also be provided as an API. Our app has a ChatBot that will intelligently help the victim by capturing his location and help him with any illness and general advice. Rahat’s Pehchan also helps in identifying missing victims.
PaperRocket - Mobile app for posting news regarding Natural Calamities Anyone in any part of the world can post news or important announcements in case of a natural disaster. If there is a new post by someone all the nearby nodes immediately fetch, thus creating a chain like network which works as long as there is someone in between two different networks seperated by a distance within Wi-Fi range. - Works with or without internet/cellular connection (especially useful when serious natural disasters like Floods, Earthquakes disrupt basic communication system like cell phone towers, fiber optical cables) - Cross platform (both iOS and android) (**NEW**) - Forms a Wi-Fi network with nearby devices even when there is no internet connection. - Spam Protection by limiting users to a fixed number of messages (**NEW**) - Works even when you don't have the app installed on your phone (**NEW**) - Works out of the box without login/registration which is really useful in emergency situations. (**NEW**) - Centralized server on Azure for keeping the posts in sync with other far away nodes
Catastrofree is a holistic solution to predict, prevent and manage natural disasters. Please read more about our project at https://catastrofree.github.io/
DUBG is a mixed reality app for efficient post disaster management. Rescuers are busy during operation, which is why we built our app based on Mixed Reality so that rescuers don’t waste time on pre-planning their routes and operating any device for communication. DUBG makes basic tasks like communication, navigation and current status monitoring easy for rescuers. Our vision is to create an Augmented Reality based navigation system enhanced by voice assistant to make the rescue drives hassle free. Using the current technology of Microsoft such as Mixed Reality and Azure Speech Recognition, we completely reimagine current Disaster Management systems.
The project is a distributed IoT-based solution, deployed in different sections / rooms of the building with the target as home / building automation solution, which acts as an early warning system and takes precautionary measures on detection of disasters, offloading crucial immediate steps from the cloud to edge, while staying connected to the cloud for data storage and further analytics. It works with an Android App (skinapp), which uses an exported tensorflow model trained on Microsoft's custom vision service (cognitive services) to predict skin diseases offline (for survivors rescued from the disaster zone), which can assist in faster skin disease diagnosis in the aftermath of disasters like floods.
When a natural disaster strikes a place, there is a need to provide shelters to people who get affected by it. We plan to create an app which will allow people looking for a place to stay can get connected to people who are willing to offer a safe place for them.
The Project Disaster Monitor aims to detect disaster through Aerial Imagery provided by satellites, it has 2 features, first is the wildfire prediction, for which we collected data for present wildfires from the earth observatory site and for absent wildfire’s, collected from screenshot sources, and trained a classifier based on transfer learning using ImageNet network ‘Xception’, and the second feature for predicting multiple disasters, same training has been done and all the data has been collected from Earth Observatory, this project can do good. if we have latest nearly real-time imagery for predicting disasters and some other things also, like spreading direction of fire and smoke, and intensity of the fire, these can be a whole lot of data which can be useful for prevention of damage that can be done through these calamities.
LimeLight is a disaster management ecosystem that aims to solve the major present-day problems in disaster relief operations, with a vision to make them effective, efficient and organized. The application has the following core functionalities: locating victims even in the absence of internet connectivity via wifi direct relay networks, real-time chat between rescuers and victims, and aerial surveillance of disaster sites via drone deployment, navigation, and real-time video streaming. Victim and rescuer locations are rendered on a map interface in the form of clusters. Rescuers can assign themselves to clusters, as well as request for reinforcements. Rescuers observing a drone's video stream can notify fellow rescuers in case a survivor is spotted in real time using the drone's GPS coordinates.
During disaster times, people struggle to find help and people who are willing to help struggle to find a legitimate source of need. Our app serves as a platform for connecting the volunteers and victims. It helps the victims post their helps and their problem is analysed and assigned a priority. Volunteers can view these posts and reach out to them. They will see the posts according to the priority of the problem. They can also search for a particular category in which they are willing to help. Once the information from the providers and victims are available the app matches them with their needs and connects them. People from the unaffected areas can see the posts and provide their help. The trusted donation portals are listed through which the users can donate to the relief funds. Our app also gives general alerts on disasters and weather. We have a bot which helps the users with all their disaster related queries.
Whenever disaster strikes, access to accurate information and the capacity to respond with life-saving assistance is critical. Our project aims to solve this exact issue by creating an end-to-end autonomous system, to provide precise information about where exactly the people are stuck, with the use of UAVs which are powered with AI and Computer Vision. Our system can efficiently distribute the drones to cover maximum area, and perform intelligent tasks such as detecting people (both visible and partially visible), whether they are able to move or not and Activity Recognition to help in optimizing the resources available.
The causes of flood can be consolidated as increase in flow of riverwater, rise in water level and increased rainfall. We have depicted the use of ultrasonic sensor to measure rise in water level. It also provides an sms alert when the threshold is crossed. the general citizens can prepare for evacuation. For practical installation purposes , the combination of pir sensors and automated rain gauges can be used in with ultrasonic sensor for better accurate results. It is adviced to maintain a database of the neighbourhood Public Organisations so that they could be made aware of upcoming disaster.
We are focusing on the predictions and management part of the theme. So, this project basically comprises of a Web App, which will show various data related to the disaster. We are utilizing social media for detections and management. As we all know, we can't really predict for sure the time and place of disasters. But, it is well known that a good contingency plan and preventive measures could save a lot of lives. And all of this with friendly User Interface. In the prototype, we will only have a working system for floods.
The idea is to design an IoT device to "reduce the risk of floods" by early prediction. We built a successful prototype device to constantly keep a check on the Water Levels and hence eliminate the need for physical human checking at river sites. The project is built on a Raspberry Pi 3.0 model B, with just 2 accessories, a Pi-camera and a portable WiFi device. The Pi-Camera is interfaced to take pictures of the water level scale installed on the sites. Making use of Neural Networks the program is trained with a suitable data-set to identify markings on the scale. The values obtained will then be transferred to any host machine at any place via WiFi and from there it is updated on a Geoportal recursively, for public access so that everyone concerned is aware all the time. If the water rises above a certain limit the authorities can be warned. We want to further explore and improvise the functionality by being able to measure the rate of change in water levels so that we can determine the level of seriousness associated with the issue. This can be achieved by using AI technology. The prototype works successfully and we implemented the same for a DAM site in Meghalaya.
People don't have a personal connection and there’s a lack of transparency when they're donating directly to an NGO. This may lead to lesser people donating, which will in turn lead to lesser funds for the relief program. Our solution is a direct peer-to-peer connection of the donor and the victim such that the donor knows who the receiver is and knows the status of his payment. Personalised and preferential allocation of resources to beneficiaries so as to maximize empathy and the human touch in the donation process.
Brahmastra comprises of two solutions to enable effective communication. The first is to leverage the limited connectivity networks using USSD and SMS to create a medium between victims and the rescue teams. Using USSD and SMS features from any phone irrespective of type, the victims can request for help thereby making the disaster relief faster. The second is a distributed peer driven LTE and rescue support network that is enabled by small portable networking infrastructure. We enabled this for small ranges using raspberry pi and enabling networking and access to data collection and enabling communication in a localized area. A collection of these devices can be networked together to increase the range of communication .
Narmada is a semi-automated system aimed at assisting post-disaster relief operations by leveraging the vast information available on online social media. Need and available tweets have vital information embedded in them, namely the resources which are needed and are being offered during an emergency situation. Narmada also performs automatic extraction of the said resources along with its location, quantity and other parameters using unsupervised NLP techniques. It performs subsequent matching of the extracted need and available tweets by computing the resource similarity between the two tweets. This eases the burden on the command centre and government agencies and enhances efficiency.
Vision: We envision to build a one-stop solution for disaster management. Our web app provides an emergency button where victims can immediately share their info(including location) and all victims are displayed on the map. This button can also be reached through WhatsApp, Messenger, etc. chatbots. The victim map will allow the govt. officials to see the victim density and hence plan the rescue missions and food supply accordingly. The chatbots also provide guidelines according to the surroundings and disaster. The family members of the victim can also search for a person using his/her image. Other than this, the app shows recent disasters and our own statistical study of disasters.
Framework for warning masses before-hand and coordinating rescue efforts after the disaster. We have 2 apps: BeSafe for Users and DisasterHeroes for Responders. Users are notified and tracked in case of disaster warning. They can also request services in general and responders will get their real-time location. In case of disaster, emergency services can register their resources and get optimum action plan based on last known location of users. On the basis of rescuers' feedback, action plan is improvised.