Gracie's Project Page

Who:  I enjoy going on outdoor adventures with my friends, whether that means skiing, camping, or walking our dogs. Some of my other favorite hobbies are reading, baking, and listening to podcasts!

My scientific/engineering interests:  Math has been my favorite subject for as long as I can remember, but I’m also very interested in data analysis and global health. I am eager to learn new things and become more knowledgeable as a woman in science, so I like situations that are intellectually challenging.

Academic goals:  I am a mathematics and quantitative analysis of markets and organizations (QAMO) double major. After undergrad, I hope to volunteer with the Peace Corps and then obtain a graduate degree.

Career goals:  In the future, I would like to find a job that connects data analysis and global health in a thought-provoking way. I’m also interested in promoting and explaining science to the public and media through science communication, so a career in science outreach sounds intriguing.

A neural network is a large class of parameterized compositional functions, which can be used for machine learning tasks such as classification, object detection, regression, and much more. In particular, when a neural network is used for the classification problem, it is trained (the optimal parameters are found) so that the network automatically assigns a class (label) to each item in the dataset. The goal of this project is to use neural networks to classify the artistic style in paintings (e.g. Impressionism, Baroque, Romanticism). Indexing large painting datasets is a laborious task, so an accurate method of automatically identifying artistic style would be a very useful tool. However, classifying paintings can be challenging since artistic style is not solely defined by visual features and especially because certain styles are similar. State-of-the-art results have been attained using transfer learning, i.e. using a network (ImageNet) that was trained on a different dataset and then retraining some of the layers in the model using the dataset of interest. This method reaches a 62% top-1 accuracy on the Wikipaintings dataset, which is a publicly available dataset containing 80,000 paintings with 25 different artistic styles. In this project, I have worked together with a group of four other undergraduate students to study the TensorFlow tutorials to learn how to use neural networks for image classification tasks. These tasks include how to train neural networks, understand convolutional networks and other network architectures, handle large datasets, and transfer learning methods. We have reproduced some of the prior results on the Wikipaintings dataset. In future work, our aim is to improve these models to reach higher accuracies, as well as use this methodology to investigate the University of Utah Marriott Library collections of paintings and photographs.