ML system development, Sep. 2020 – Oct. 2020

I joined a short-term project focused on developing an efficient method for generating machine learning training data. In machine learning applications, preparing a sufficient volume and variety of data is often a critical prerequisite for success.

While data augmentation is a common approach, our project required a method that could generate even greater diversity along a set of prescribed features. To address this, we developed a Proof-of-Concept (PoC) program that generates bulk data with coherently randomized local variance.

A key outcome of this project was the creation of an auxiliary tool, MarblingPy. This open-source Python library is an implementation of the image rendering methods introduced by Shufang Lu’s research group in their 2011 paper, “Mathematical Marbling.” The tool was developed in collaboration with Taktpixel Co., Ltd..