Katja Della Libera is an undergraduate student at Minerva Schools at KGI, where she first encountered coding as well as complex systems. Previously, she was part of many programs like the United World Colleges and the Congress-Bundestag Youth Exchange. She has passions ranging from musical theater to sustainability, and just recently discovered her love for computer science and programming. Katja hopes to make the topic more interesting and less scary to many people in fields that could benefit from it by applying it to various areas of interest.
Project: Reading Bar Charts Using Machine Learning
The goal is to train a neural network to read charts from pictures, similar to what is already possible with text recognition. The subproblem I am focusing on for the project is reading bar charts.
Main Results in Detail
I created a function that creates a bar chart from the image of a bar chart. It uses two neural networks, one reading the highest bar of the bar chart and one reading the relative heights of the bars. The biggest challenge was in the diversity of different visual representations of bar charts with mesh, background colors, etc. Over the course of the project, I generated more and more data to expand the types of graphs my network was able to read, but there are still several problems to fix.
There are many expansions and improvements to what I have done in the two weeks at the Wolfram Summer School. I would focus on fixing the existing bugs and eventually make the network a subset of a larger network that can identify graphs, align them in the right direction and read them independent of the type.