4303 research outputs found
Sort by
Breaking Hearts: AI Approaches to a Trick-Based Card Game
No full textA significant portion of artificial intelligence research has been devoted to playing games. Games allow simple, idealized environments to test and better understand the functions and weaknesses of various AI approaches. Although significant research has been done applying AI to well known games like Chess, Go, and Poker, other games are underrepresented. In our project, we set out to create multiple appraoches to playing the card game Hearts with the goal of not just creating the strongest possible agent, but also ones that are fun to play against
QUEENIE
No full textTwo sisters are forced to stay at their quirky Aunt Sherri\u27s house - but she\u27s acting a lot weirder than usual..
Thinning is Winning: Using Data Thinning as an Alternative Approach to Sample Splitting
No full textIn this paper, we explore data thinning, a method in which a variable is decomposed into two or more independent variables following known distributions. Data thinning relies on data coming from known distributions with sufficient statistics, a set that spans the entirety of the exponential family (Normal, Poisson, Binomial, Gamma, etc) as well as distributions with varying support such as the Uniform distribution. We first perform a literature review to present current knowledge regarding thinning. We then proceed by examining data thinning under ideal conditions, before examining its performance when these conditions are not met. Via simulation, we demonstrate the potential benefits of data thinning’s application to the model validation process as an alternative to sample splitting. Finally, by performing data thinning on SO2 pollution data from across sixty cities in the United States, we argue for the use of data thinning as an alternative method for performing model validation and inference with real world data in situations when sample splitting fails
A Trip Through the Zoo of Infinite Abelian Groups
No full textIn this paper, we explore infinite abelian groups and their properties. We begin with a self-contained proof of the fundamental theorem of finitely generated abelian groups. Next, we define divisible groups and prove a structure theorem for them. Finally, we switch gears and talk about the p-adic numbers, exploring them both algebraically and topologically
Crude Consequences: A Comparative Analysis of Oil Spill Impacts on Aerial Versus Submerged Species in Saltmarsh Ecosystems
No full textExploring Multiple Paternity and Female Choice: A Strategy to Enhance Genetic Diversity and Population Viability
No full textReinforcement Learning in Monopoly Through Three Q-learning Variants
No full textMonopoly is a complex game with many random variables and complicated decision processes. Its complex dynamics, involving property acquisition, resource management, and player interaction, make it an ideal testbed for artificial intelligence (AI) techniques. In addition to the traditional fixed-policy strategy, there is also a rise in exploring reinforcement learning algorithms in games like this --- among which Q-learning is a simple and adaptive example.
This project builds and compares three types of Q-learning agents for monopoly: the basic Q-table approach, approximate Q-learning, and deep Q-lambda learning. The basic Q-learning approch is simple to code and quick to train, but it may not be able to capture the complex dynamics of the game; the deep Q-lambda agent considers complex situations in the game and long-term rewards, but takes exceptionally long to train; the approximate Q-learning agent lies in-between, considering the complex dynamics but not long-term rewards, and trains in a moderate amount of time. With these variants we aim to explore the balance between efficiency, simplicity of code, and the ability to adapt to the large state-action space and the need for long-term strategic planning.
Our agents were trained using gamescomputersplay/monopoly, a monopoly game simulator available on github. Our agents played fixed-policy players under simple and complex game conditions. The agents use a reward function that balances its property (land, houses) and money. Each agent was able to stablize and perform better than a random action agent. For demonstration purposes, the deep Q-lambda agent was exported using Open Neural Network Exchange and incorporated into intrepidcoder/monopoly, a runtime JavaScript-based monoply game board that allows you to play with our agent
Macroeconomic Drivers of Sectoral Volatility: Evidence from GARCH-Family Models
No full textWe investigate the impact of macroeconomic indicators—inflation, interest rates, and unemployment—on stock market volatility across Energy, Financials, Healthcare, and Information Technology sectors from 2001 to 2024. Our analysis combines panel regressions with GARCH-family time series models, including a GARCH-MIDAS extension that accommodates mixed frequencies between daily stock returns and monthly macroeconomic data. Both panel estimates and GARCH specifications indicate that inflation exhibits the strongest impact on realized volatility, particularly in the Financial and Information Technology sectors, while unemployment effects comparatively weak. Sub-period analysis around the 2008 financial crisis and 2020 pandemic reveals heightened sectoral sensitivity to inflationary shocks, emphasizing the importance of industry-specific risk factors during market turbulence. While the GARCH-MIDAS framework theoretically offers finer decomposition of volatility components, its predictive improvements are limited
