This project explores how adversarial behavior and strategies develop when two agents with the same neural network “brain” play a game against each other in a simulated environment.
This project was inspired by Open AI's paper examining the use of tools by agents playing a hide and seek style game. The agents were created using Unity Machine Learning Agents (MLA) utilities.
The world of AI is growing at an exponential rate in the real world, but what about the virtual world?
"Blu" is an agent created to see what is possible with deep learning in the world of GTA V. Even if that goal is to punch as many people as possible ...
The People's Republic of China has been publishing “fake news” in Taiwan in an effort to delegitimize Taiwanese autonomy and strengthen their presence in the region.
This project features a model that takes a news article written in Traditional Chinese and returns the probability that the author is Taiwanese or mainland Chinese. These sorts of models can help prevent the spread of misinformation in the Taiwan-China discursive battlespace.
In this project, an agent is trained to develop a betting strategy for Blackjack. This agent is trained using reinforcement learning in Open AI Gym.
This project explores the state-of-the-art in neural puppetry to to animate a still image of Dr. Watson speaking a line from a machine-generate Sherlock Holmes story.
The final product combines a text generation model, a voice-cloning model, and a first-order motion model.
This project features an neural network that models and predicts the type of play a collegiate football offense will run in a given situation. This model could help Army's defensive coaches choose a defensive plays that minimizes the opponent's chance of success.
The model was trained against all 2017 Army v. Oklahoma games to demonstrate the model's accuracy in a realistic scenario. This project also includes a simple GUI to easily enter the model's input.
We know that neural networks can create highly realistic photos, but can they make music? This project experiments with music generation using regressive neural networks.
This project uses Generative Adversarial Networks (GANs) to predict sporting event upsets. We compare the efficacy of this approach to more traditional machine learning approaches and baseline sports betting approaches and results show that this approach outperforms the baseline and machine learning approaches.