Can Transformers Learn to Strategize?

TicTacGPT for playing simple board games

Charlie O'Neill
Towards Data Science
27 min readSep 8, 2023

--

Photo by Jon Tyson on Unsplash

Although most board games appeal to the use of convolutional neural nets or other geometrically inspired architectures, the fact that we can actually represent board states as strings begs the question of whether transformers can naturally be applied to the game. Here, we’ll see if we can answer this in the context of a simple game…

--

--

Researcher at Macuject, ANU. Interests include integration of deep learning, causal inference and meta-learning. Twitter: @charles0neill.