IEEE Conference on Computational Intelligence in Games
NYU Brooklyn 22 to 25 August 2017
Conference archive
Conference Details / Program
A browsable record of the 2017 week. Move through the four days, or open a single paper session to see its chair, time, scope, and place in the conference.
The printed 2017 program included individual paper titles, author lists, and a PDF schedule. This archive preserves the session structure, chairs, times, and themes without reproducing those documents.
Two paper sessions and the first competition block
Planning methods took center stage: tree search in a card game, map aware balancing for location based play, difficulty adjustment, and the challenge of giving several agents distinct roles in one world.
Work on player modeling moved from raw behavior to useful signals: action features, strategic depth, simulated dexterity, and the facial expression of flow.
Teams introduced the year's entries in a fast sequence of talks, moving from real time strategy to first person play, collaborative puzzles, and turn based combat.
Four paper sessions, from nominees to late learning
The nominated papers covered a text adventure agent, a fuzzy system for public goods games, a hybrid of deep learning and neuroevolution, and a stronger Hearthstone bot.
Social non player characters, an agent for Hanabi, human like learning, and hierarchical task networks for Minecraft.
Partial observability, macro management in StarCraft, resource gathering, and coevolution in Pac Man.
Visual Doom and a puzzle game served as test beds for reinforcement learning and generalization.
Short papers, procedural content, and competition talks
A compact block of current work: neural and evolutionary heuristics, deep Q networks for fighting games, live service optimization, churn prediction, FPGA simulation, episodic control, player psychology, and opponent modeling.
Level generation for physics games, mixed initiative dungeon design, evolving maps for Cube 2, and automated game design learning.
The second competition block covered general game playing, level generation, commercial game data, and two very different tests of control under uncertainty.
General game AI, generation, and the closing session
Rolling horizon evolution, diversified heuristics, temporal difference learning, and real world physics with macro actions.
Rule generation, sprite generation with deep convolutional networks, multi layer level representations, and pattern based Angry Birds levels.
Communication in pursuit domains, the Showdown competition, adaptive mobile gameplay, and a test of the no free lunch idea in general game playing.
The archive lists the shape of the week and the people who chaired each session. The proceedings remain the place to find the papers themselves.