“Tread softly because you tread on my dreams,“ W. B. Yates.

Our footprints matter, at the individual, and collective level. Our students know and care about it, and more will do so if they can measure, analyze and understand the implications of what, how much, and when they consume, save, and how their choices make a difference. We cannot think of a better way to deploy this idea than in one of the undergraduate houses. This proposal aims to “measure Mather” where the faculty deans have for years been emphasizing a vision of sustainable living. Using sensors of different types, through the seasons and semesters, this proposal has the goal of measuring consumption at multiple levels, and educating our students as they live and learn about their community, and make informed choices about how to change for the better in a sustainable community.

How much energy is used and when? How much water is used and when? How does the temperature vary in different spaces? What are the concentrations of carbon dioxide, humidity in our rooms and public spaces and how do they vary? How much food do we eat? Waste? How many people go and in out of Mather? How can we implement differential privacy protocols and collect data of this kind to get at aggregate statistics while maintaining individual anonymity? What types of data science tools combined with mathematical models can we deploy to understand the coarse and fine grained patterns in the data? And how can we begin to use the measurements and analyses to debate changes in policies while informing decisions about resource allocation for material changes and consumption and saving habits towards a mindful approach to community living?

These are the questions that the proposal aims to answer using the lens of science and technology, while engaging the entire community that will be affected by any decisions.

Measurement requires tools that are cheap, accurate and robust, and easy to deploy and collect data from. They also need to encode protocols that protect privacy. TinyML - the ability to combine hardware, algorithms and software in low power always-on devices that also carry out basic data analytics provides a natural solution to this at scale. Using ML at the endpoint will allow us to dramatically filter out unwanted data and only retain relevant information for more intelligent processing.