Inverse Attention Agents for Multi-Agent Systems

Authors: Qian Long, Ruoyan Li, Minglu Zhao, Tao Gao, Demetri Terzopoulos

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments in a continuous environment, tackling demanding tasks encompassing cooperation, competition, and a blend of both. They demonstrate that the inverse attention network successfully infers the attention of other agents, and that this information improves agent performance. Additional human experiments show that, compared to baseline agent models, our inverse attention agents exhibit superior cooperation with humans and better emulate human behaviors.
Researcher Affiliation Academia Qian Long UCLA EMAIL Ruoyan Li UCLA EMAIL Minglu Zhao UCLA EMAIL Tao Gao UCLA EMAIL Demetri Terzopoulos UCLA EMAIL
Pseudocode Yes Algorithm 1: Inverse Attention Algorithm
Open Source Code No The paper does not contain any explicit statement about releasing source code for the methodology or provide a link to a code repository.
Open Datasets Yes We demonstrate the effectiveness of our approach through a series of experiments adapted from the Multi-agent Particles Environment (MPE) (Lowe et al., 2017). All of the environments used in our experiment were built on top of MPE (Mordatch and Abbeel, 2017; Lowe et al., 2017).
Dataset Splits Yes 70% of the dataset is used for training, 10% of the dataset is split into a validation set, which is used for early stopping, and 20% of the dataset is used for testing.
Hardware Specification Yes Hardware specifications: All experiments are run on servers/workstations with the following configurations: 128 CPU cores, 692GB RAM; 128 CPU cores, 1.0TB RAM; 32 CPU cores, 120GB RAM; 24 CPU cores, 80GB RAM, 1 NVIDIA 3090 GPU.
Software Dependencies No The paper mentions several MARL training schemes like MAPPO, IPPO, MAA2C, and To M2C* as baseline methods but does not specify specific version numbers for underlying software libraries (e.g., Python, PyTorch, TensorFlow) that would be needed for replication.
Experiment Setup Yes Table 7: Hyperparameters for the gradient field; Table 8: Hyperparameters for the agent training; Table 9: Hyperparameters for the inverse network