Inspiration
Everyone wants to make sustainable choices, however, this decision rarely remains at the forefront of an online shopper's mind. We developed an application to inform the user of the sustainability of different fashion companies to remove barriers toward sustainable decision-making. Not only do we aim to increase sustainable habits, but we built our algorithm using hyperdimensional computing, a sustainable approach to machine learning due to the low computational costs. We believe that the tools to achieve sustainability should also be sustainable in themselves.
What it does
When you traverse fashion sites, our Google Chrome extension displays a sustainability rating by performing hyperdimensional computing on different sustainability features such as their carbon footprint and the fashion transparency index. We parse data from different sites that provide this information and feed our input into the HDC algorithm, which then concludes the output given our previously encoded relationships.
How we built it
Google Chrome Extension: The extension works as a basic React app that accesses the page URL the user is currently on and uses that to find data about that company's sustainability. We gathered the data from 'https://directory.goodonyou.eco/‘ which provided information on the sustainability of different fashion brands.
Hyperdimensional Computing Algorithm (HDC): HDC represents a novel way to approach machine learning which tends to be less computationally expensive because you are only using boolean algebraic operations. First, we separated the idea of sustainability into multiple features and then assigned hyper binary vectors of dimension 10000 to each state of each feature. Using dictionaries, we made sure that the table with the hyper binary vectors had the same consistent hyperbinary vector for each distinct state for each feature. We use the majority bit rule row-wise between each state of sustainability to create a "scene". Each row of our sustainability table represented a scene, and then we XORed these scenes with the output list which represented the conclusion of whether or not the company was sustainable. Finally, we use majority bit rule to "bundle" together this Scenario for an Experience Vector. This is our training portion. Once we have this experience vector, we "query" this Experience vector with our given scene and compare the result's hamming distance with all the vectors in our current output dictionary. We return the value with the smallest hamming distance, which represents the most similar vector.
Challenges we ran into
For the chrome extension, we ran into issues when we wanted to use npm packages (i.e. Material UI to help style the application) and so we needed to look for external packages for Chrome Extension like webpack to bundle all our files together. When researching this implementation, we realised we could just use React instead of Vanilla JS so we switched over. With the web scraping, there were different issues with the webpack configuration and chrome extension requirements so we ended up having to use sample data to demonstrate our proof of concept.
One of the largest issues existed within the integration between the chrome extension which had to be built in javascript and the hyperdimensional computing algorithm which was written in python. Additionally, hyperdimensional computing can get complicated with the different various operations so there was a lot of debugging involved and it doesn't represent the full capabilities of HDC because of the time constraints involved. And I think the algorithm could be more efficient through another method rather than turning it into a dictionary to keep the same consistent value throughout.
Accomplishments that we're proud of
We are proud of taking an unconventional route towards an idea where we were able to develop a proper proof of concept that demonstrated the main points of the product.
What we learned
We learned how to perform rapid ideation by going through various ideas but then committing to an MVP of one main idea.
What's next for Meta-Sustainable Fashion Algorithm
We want to focus on proper integration of the hyperdimensional computing algorithm onto the chrome extension. Furthermore, we aim to add more features of sustainability within the rating.
Log in or sign up for Devpost to join the conversation.