Deep Learning

Selecting hyperparameter values for a neural network can greatly improve performance. With many potential combinations of hyperparameter values, however, it can quickly become intractable to test combinations manually. R’s tfruns package helps implement this search and identify the best values.

I develop a framework to flexibly incorporate both time-varying and static variables into a river temperature model with over 4 million observations.

Advances in image super-resolution have led to the ability to generate strikingly photo-realistic images from low-resolution inputs. Inspired by these models, we create high-resolution maps of future wildfire risk, incorporating satellite observations, weather data, and climate models.

Advances in image super-resolution have led to the ability to generate strikingly photo-realistic images from low-resolution inputs. Inspired by these models, we create high-resolution maps of future wildfire risk, incorporating satellite observations, weather data, and climate models.