We extend pymarl2 to pymarl3, equipping the MARL algorithms with permutation invariance and permutation equivariance properties. The enhanced algorithm achieves 100% win rates on SMAC-V1 and superior performance on SMAC-V2.

Official implementation of "RE-MIMO: Recurrent and Permutation Equivariant Neural MIMO Detection" paper.

Graph neural network autoencoders for jets in HEP

EquiFusion (ECCV 2026): kinematics-agnostic stochastic 3D human motion prediction via permutation equivariant latent diffusion. Multidataset training unlocked and generalization to unseen kinematics (cross-dataset); zero-shot prediction from occluded input and targeted limb generation.

Covariant Compositional Networks Library is an easy-to-use and efficient implementation of Covariant Compositional Networks (CCNs) with TensorFlow and PyTorch's APIs based on a shared common C++ core.