So, What Exactly is Machine Learning?: Inside a CEE Capstone Project
By Jill Dressler ’18 and Rachel Galowich ’18
Machine learning is a 21st century buzzword, but what does it really mean? When we sat down in Professor Saurabh Amin’s office to first begin discussing our senior capstone project, machine learning sounded like a concept from another planet, a field that only computer scientists could enter. Yet, here we are, about two months deep into a fascinating capstone project that applies machine learning theory to civil engineering applications in transportation. And we are thriving.
As Course 1 undergraduates, we have to complete a senior capstone in order to graduate. It serves as an opportunity to get a taste of a semester long research project, perhaps to inspire some of us to on to get our PhDs or write a thesis for a master’s degree. I’m Jill Dressler, and I’m writing this blog post with my colleague and best friend, Rachel Galowich. We are both doing our capstone project under the advisement and support of Professor Saraubh Amin and Doctoral Candidate Jeffrey Liu.
We hope that this blog post and the few that follow it can give you a bit of an inside look into our project. Here is a little bit of background information:
The Massachusetts Department of Transportation (MassDOT) has several hundred cameras scattered across Massachusetts roadways. They oversee traffic flow, infrastructure quality, and traveler safety across thousands of miles of streets and highways…and they are recording it all! The hundreds of cameras stream hundreds of thousands of images to the Highway Operations Center every day. Dedicated MassDOT employees monitor the roadways and respond to events as needed. The current system faces several technical limitations, including delayed response time to weather events, heavy traffic, and vehicular accidents.
So basically, we have all of these images being processed into a database, but we have not even broken the surface of their potential. Google has a software that can place labels on these images (think: “asphalt,” “snow,” “bridge,” etc.). From here, Professor Amin has invited Rachel and I on board to help use the labeled images to develop classification and prediction methods for future data. This may sound convoluted, so let us try to simplify it.
There are two scenarios that may make this a bit clearer. First, imagine that a police dashcam takes a picture on a highway and we need some way to localize the image. Can we use the existing image database to organize the data in such a way that we can quickly assess where the image was taken based on label similarities? Second, imagine it’s a Tuesday at 5:33PM and you are sitting in dead-stop traffic on the freeway. Was there any way to predict this seven-minute span would have significantly worse traffic flow than the previous seven minutes? These are simplifications of the questions we hope to answer.
Stay tuned for our next blog where we will go into a little bit more detail about how we are employing machine learning techniques (but don’t worry, we won’t go all super-nerd on you!). Thanks for reading!