Research and Projects

Neon Waves: A Model For Computer Generated Simulation of Bioluminescent Ocean Waves

The field of natural world simulation has been revolutionized and has grown exponentially in the past decade, and sophisticated algorithms have been developed to simulate everything from elasticity to fluid simulation. However, the field is still relatively young, and there are many natural phenomena that have not yet perfected or even attempted in animation. Once algorithms have been perfected to replicate a specific phenomenon like fire or fluid, these tools can be used by animators for years to come to create compelling and realistic depictions of the real world. The goal of my project is to study the patterns of bioluminescent bacteria in ocean waves, and to recreate that phenomenon in the effects software, Houdini. By deriving the equations to model the luminescence of the bacteria, I plan to model the waves themselves using a FLIP solver, and then give a density to each particle which represents the amount of bacteria in that particle. The differential equation for light activation and quenching will then be used to light up the particles, with the wave acceleration serving as a stimulant.

ESGF in the cloud: A Community-driven Effort for Scalable Data Access and Analysis

The Earth System Grid Federation (ESGF) is an open source, international collaborative effort providing a robust, distributed data and computation platform for Earth System Science, notably used in disseminating Coupled Model Intercomparison Project (CMIP) model output from a variety of climate modeling centers. Modeling centers around the world have served their data through a network of public-facing nodes deployed by the ESGF consortium, where data can be made more easily discoverable, comprehensible, and downloadable. In light of the rapidly growing climate data archive and the community’s increased interest in scalable analytics (e.g. multi-model analysis, machine learning and downscaling applications), we turn our attention to the concept of bringing analysis to the data as opposed to data to the analysis. In this presentation, we discuss a major milestone that the ESGF team, led by the Geophysical Fluid Dynamics Laboratory (GFDL), has accomplished as part of an effort to deploy a more dynamic, scalable, performance-based, and cloud-optimized infrastructure (using AWS Elastic Kubernetes Service). We incorporated ESGF’s containerized software stack to publish, track, and curate NetCDF datasets in the cloud. In addition, we utilized Amazon’s S3 cloud storage and mounted it as a POSIX-like file system (via Goofys) to make accessible nearly 1 petabyte of CMIP6 datasets. We have also successfully deployed multiple retrieval, cataloging, and on-demand, large-scale data analysis packages that utilize S3, including Pangeo-inspired tools, Unidata’s THREDDS data server with OPeNDAP, and Synda. These applications demonstrate some of the ways researchers can access and analyze datasets in the cloud, opening the door for many potential technological and scientific innovations to be realized through community collaborations.

Team Rocket: Investigating Inter-Agent Communication and Machine Learning in Rocket League Bots

Artificial intelligence is a growing field in the realm of video games, however many intelligent agents are still lacking in the area of communication. This paper explores various methods of communication and teamwork among artificially intelligent agents in the game Rocket League from simple inter-bot messaging to a shared controller for several bots. It was hypothesized that the ability to communicate between agents would make teams more successful in matches of 2vs2or more. This hypothesis is supported by the results of this study

Analyzing Objectivity in Film Maturity Ratings

The maturity ratings of films given by the Motion Picture Association of America have long remained mysterious to the public. This paper seeks to demystify these ratings somewhat by analyzing the ability of various classification models to predict the ratings given to a test set of films. Because the ratings are determined by a group of individuals with a vested interest in the outcome, the prediction is that models will not be particularly successful at determining ratings from the scripts alone. This prediction is supported by the results of this project.