Bennett Dell Norman
Howdy! I am Data Engineer at Catalyst Coop interested in using data to support urban planning decisions and our clean energy transition. Here is a link to my resume. This website showcases some of my projects. Enjoy!
Affordable Opportunity
Description
I used data from Opportunity Insights and NYC Open Data to map and analyze the economic mobility of neighborhoods with new affordable housing stock. Since 2014, a quarter of new affordable units in NYC are in moderate to very high economic opportunity tracts, the rest being in very low and low opportunity tracts. The Github readme contains a description of my methods and findings. You can explore the data on this kepler.gl map.
Collate | NYC CEQR Review Software
Description
Collate was an urban planning startup project for Cornell Tech’s startup studio course. Collate aimed to support real estate developers during the research-intensive rezoning environmental review process called CEQR by providing data and models required by CEQR.
This project taught me a lot about the urban planning procedures in NYC, and it introduced me to the world of urban planning tech. Here is a link to our pitch deck.
CA Property Tax Map
Description
I contribute to the Property Tax map, which scrapes county tax assessors' websites to create an open-source data set of property tax data for almost all parcels in California. This data is important for researchers, policymakers, and individuals to understand California's outdated property tax policies.
Economic Opportunity Analysis
Description
I am inspired by the work at Opportunity Insights . They use data from multiple censuses to understand how the income of adult children compares to their parent's income. Using this data, they can accurately describe economic mobility between gender, race, and location. I performed my own analysis using their publicly available data. Here is a link to the git with my analysis and code.
Refering Expressions
Description
Semantically Aware Objective Functions for Referring Expression Tasks is the culmination of a years work in PhD classes CS 6670 Computer Vision and CS 6784 Advanced Topics in Machine Learning taught by Professors Barath Harihardan and Kilian Weinberger respectively. I partnered up with two of my friends to improve on the Referring Expressions task. The task is as follows: given an image and a region of the image, generate an unambiguous expression. We greatly improved on our baseline by comparing the sementaic meaning of expressions during training.
Abstract
In a previous work, we proposed a set of objective functions that can be used for any text generation task. These functions utilize recent progress in high-quality word embeddings to gain an awareness of semantics. In a holistic review of our past and current work, we explore their efficacy and their applicability to the Referring Expression task and Language Generation tasks in general. Link to the paperAudio Feature Extraction
Description
For our Artificial Inteligence Practicuum, our team used Google Magenta's audio autoencoder to extract high quality features of a large music collection. We used these feature vectors to find an artist's "average song", perform analagous reasoning tasks such as Old Man - Neil Young + The Beatles = Julia and find the most similar and dissimilar artists. Here is a link to the report.
Architecture
My interest in generative design and computer science lead me to the Jenny Sabin Design Lab at the end of my sophomore year. The lab encourages interdisciplinary collaboration among a small group of undergraduate architecture, computer science, electrical and materials engineers.
RoboSense
I worked closely with two architects developing software that permits architects to initialize and control dynamic fabrication environments with a 6 axis robot and arduino end effectors. We presented our accepted paper at Acadia 2018. This is a photo of a MOMA PS1 spool weaving end effector that uses a stepper motor and load sensor to maintain consistent tension.
Kirigami Simulation
I produced Unity simulations of kirigami geometry designed by the architects of the lab. We published a paper in SimAUD 2017 that explores the possibilities of kirigami geometry — folding with the addition of strategically placed cuts and holes — through simulation and kinetic and adaptive architectural assemblies. Download the paper here.
