While there isn't a single famous academic "Deep Paper" by that exact title, the phrase typically refers to research in using Super Mario Bros. (NES) as a primary benchmark for AI agents . Core Research Themes
: This research paper discusses using Deep RL to tackle the vast state spaces of NES titles, noting that in an average Mario level, a character can occupy thousands of different x-positions across multiple timesteps. Super Mario Bros NES
Research involving Super Mario Bros. on the NES often focuses on training agents to navigate complex environments using only visual input. Key papers and projects include: While there isn't a single famous academic "Deep
The "depth" of the NES original is also frequently discussed in the context of its legacy: Research involving Super Mario Bros
: High-grade, early-print "sticker sealed" copies of the game have sold for record-breaking amounts, including a notable $100,000 sale . Building a Deep Q-Network to Play Super Mario Bros
: Many implementations, such as those found on Paperspace , detail building Double Deep Q-Networks to teach agents how to clear Level 1-1 by updating "Q-tables" based on reward functions.
: Projects like ArvindSoma's A3C build upon the foundational paper "Asynchronous Methods for Deep Reinforcement Learning" to train agents specifically for the NES environment. Technical Context of the NES Original