ML system development, Sep. 2020 – Oct. 2020

We had joined a short-term project out of several machine-learning involved tasks where we seek for an efficient way of training data generation.

In a field, it is often essential to prepare a sufficient variation/amount of data whenever a methodology of machine learning is applicable.

While Data Augmentation is well known approach among such, for our purpose it was required to have more diversity along with a prescribed feature, we developed a program aiming for PoC that generates a bulk of data whose local variance is coherently chosen to be random.

The project gave rise to an auxiliary tool, MarblingPy, a Python implementation of the image rendering methods introduced by the research group of Lu, Shufang in “Mathematical marbling.”, 2011., under the collaborative work of Taktpixel Co., Ltd..