IEEE Conference on Computational Intelligence in Games
NYU Brooklyn 22 to 25 August 2017
Conference archive
Conference Details / Competitions
Twelve contests turned the conference into a working benchmark. Agents planned, learned, generated, collaborated, and competed across very different games.
12 competitions in view
Jialin Liu, Diego Perez Liebana, Simon M. Lucas, Ahmed Khalifa, and Julian Togelius
One controller, many games it has never seen. The 2017 edition added a learning track and a rule generation track alongside level generation.
Kyung Joong Kim and Seonghun Yoon
Bots managed economies, armies, and incomplete information in a real time strategy classic. The contest pushed on uncertainty, speed, and unit control.
Jochen Renz, Julian Togelius, Lucas Ferreira, Matthew Stephenson, and XiaoYu Ge
Generators produced new levels judged for human fun and for difficulty. The contest ran jointly with IJCAI 2017, with remote entry allowed.
Rui Prada, Francisco S. Melo, and João Dias
Two agents, two shapes, and one physics puzzle. The test was collaboration under motion constraints, with a single AI track for teams that wanted the planning problem without the coordination layer.
Santiago Ontañón
A small scale strategy game built for planning research. Agents could use a forward model, which made search and plan evaluation practical without a large engineering team.
Tim Atkinson, Hendrik Baier, Tara Copplestone, Sam Devlin, and Jerry Swan
Language, exploration, and partial knowledge. Agents worked through a graded set of adventure games, with credit for both the score and the amount of domain knowledge they avoided.
Ruck Thawonmas
FightingICE provided the arena. Standard and speedrunning leagues tested robustness across three character types, with a scoring system borrowed from Formula 1.
Wojciech Jaśkowski, Michał Kempka, and Marek Wydmuch
Agents played from pixels in a multiplayer deathmatch. ViZDoom exposed the screen buffer as the controller's only view, inviting deep reinforcement learning approaches.
Piers R. Williams, Diego Perez Liebana, and Simon M. Lucas
A contest about what an agent cannot see. Entrants controlled either the chased character or the chasing team, with separate tracks for each side.
Scott Lee and Julian Togelius
Team building and move selection in a Pokémon style battle system. The game offered hidden information, random outcomes, and a large space of possible teams.
Tobias Mahlmann
Five heroes, one team, and no backchannel chat. Bots had to coordinate through the game's limited signals while a custom mod let external processes control each hero.
Boram Kim, Du Mim Yoon, Eun Jo Lee, JiHoon Jeon, Kyung Joong Kim, Sang Kwang Lee, and Sung Il Yang
Prediction from commercial game logs, including churn and engagement after a payment model change. Entrants worked with data recorded by NCSOFT, a major South Korean game company.
No competitions match that search. Try a different word or return to all contests.
The 2017 entry forms and rule sets are closed. This page records the contests and the questions they posed, not the current state of each competition.