** Overview of project:
Statistical modeling experiences waves of popularity. Energy-based models (EBMs) have ridden these waves twice in the last two decades, first gaining popularity in 2003 why, then quickly becoming overshadowed by deep neural networks in 2012. EBMs have recently garnered renewed attention as new approaches to training have shown their ability to perform well against strong generative adversarial networks (GAN) (Song, 2019; Song, 2020). Aside from their performance, EBMs provide a unique approach to capturing uncertainty in high dimensional spaces. Real world applications often do not yield clean deterministic outcomes, presenting a need to make predictions in the presence of uncertainty. This is an area where EBMs excel. This paper explores the architecture of energy-based models by tuning the model, benchmarking against strong GANs on a variety of datasets, and experimenting on applications of object detection and image restoration.