Recognize the Right Problem

Now, one fine summer of 2014, Ahmed went back home to Kashmir. And I'm not sure if you know about this, but Kashmir, India, had one of the largest floods that year. A flood which they hadn't seen in several decades. We tried very frantically to reach out to him. We would see the news, talking about how many hundreds of people ended up losing their lives. And we couldn't reach out, we couldn't connect with him. And it was very, very scary. Ultimately, we did end up connecting with him to figure out that his family and him were safe after the flood. However, when he came back to Stanford, he talked about so many stories related to the Kashmir flood, and it was pretty mortifying. It was a disaster in all sense. There was complete chaos after the disaster. You would have no sense for the emergency rescue to figure out where they should actually go, who should they actually rescue. You had thousands of people abandoned on the rooftops for several days and you have hundreds of people either drowning or being washed away by the flood. And we wondered, why is this still a problem and why in the 21st century, why is nobody thinking about it? And what can we do to help, just being graduate students here? A little while later, we got the opportunity to work in a class project together. It was the machine learning project at Stanford, as well as the probabilistic earthquake engineering class at Stanford. We took that opportunity to figure out, can we actually solve that problem we saw in the summer of 2014, and what can we do to reduce that chaos? The goal there was just let's find a way to figure out what exactly the problem was. And the problem we tried to reduce resolve was can we provide a real time granular understanding of which buildings and which people are in what state of collapse during a disaster? So that was our problem statement. Our backgrounds of computer science, structural engineering and earthquake engineering led to it being focused mostly on seismic initially. And we were very happy with the results. It was surprising because it wasn't done before, but there's a lot of physics and science already there in structural. In earthquake engineering. All we did was sort of understand the features present there and really convert that in a more granular and real time solution. We're happy with the algorithm we created, but what we were more pleasantly surprised was that when we presented this algorithm in the CS229 machine learning fair, a lot of professors came up to us, a lot of investors came up to us and said that what you have is applicable and you can create a real product here.