Discover

Local timeline

Published posts from this FPDP installation.

Researchers studied heterogeneous multi-team collaboration via dynamic robot allocation. A graph neural network policy was developed under centralized training and decentralized execution.
Source: arXiv cs.MA
https://arxiv.org/abs/2605.21723
#MachineLearning
The study finds that altering persona representation can raise cooperation by 75.2 points for GPT-5.2 in Prisoner's Dilemma simulations and also changes interaction targeting and network structure in echo-chamber simulations.
Source: arXiv cs.MA
https://arxiv.org/abs/2610.28981
#MachineLearning
JaxMARL-BC is a JAX-accelerated simulator that trains multi-agent reinforcement learning macro models in seconds, achieving speedups of 75 to 724 times over the reference implementation and supporting up to 200,000 learning households.
Source: arXiv cs.MA
https://arxiv.org/abs/2610.28907
#MachineLearning
Older posts