COSA awarded several micro-grants to technologists working on projects dealing with ML5, an open-source machine learning library. Yining Shi and Joey Lee refactored the Neural Network feature in ml5 in hopes of making the API “as friendly and maintainable as it is efficient and powerful.” This intervention aims to make the software easier to maintain in the long run. Lydia Jessup also set out to widen the ml5 Neural Network’s accessibility by creating a walk-through tutorial on how to prepare a training dataset, aimed at audiences that may not necessarily come from statistics or computer science backgrounds.

Ellen Nickles’ proposal similarly addressed the issue of training models: in her Data and Model Provenance Project, she researched a number of contextual considerations surrounding the pre-trained models that currently exist in the ml5.js library. By collecting more detailed descriptions of their training data and original use context, this information—which Nickles intends to publish on the “Resources” page of the ml5.js website—can help users evaluate the exigencies of existing data with a deeper understanding of its potential quirks and biases. Aidan Nelson’s project similarly seeks to streamline and document existing resources around machine learning: Nelson proposed to document and update the open-source ML-Agents Toolkit, an addition that enables users to train models through reinforcement learning, a specific modality that enables visual feedback to be integrated into the training process.