Plot Generating AI

They say pictures are worth a thousand words, and in the experimental research the equivalent might be a figure is worth a thousand data points. Our objective is to build a generative adversarial network (GAN) that can generate "good" plots. Where a "good" plot is defined as displaying correlation, physically interpretable and meaningful, and following the principle of parsimony. We propose three metrics for defining what makes a "good" plot. Then with an initial random guess we can optimize the initial guess until we arrive at the nearest "good" plot; this plot is defined as a potential candidate. Iterating many times we can create a sample of candidates. The sample is combined with actual plots gathered from literature and then used in the creation of the generator, a convolutional neural network, and the discriminator, a deconvolutional neural network.