graph-attention

A hierarchical multi-agent reinforcement learning framework combining graph attention and dynamic role assignment improves tactical coordination and win rates in air combat.

This paper introduces Power Law Graph Attention (PLGA) and the PLDR-LLM architecture, an exact generalization of scaled dot-product attention using input-generated bilinear operators. It presents theoretical results including an inference-collapse theorem, empirical stability measurements, and machine-checked proofs in Lean 4.

This paper proposes LTGA, a graph attention layer that learns a per-edge Tsallis entropic index to interpolate between heavy-tailed, softmax, and compact-support attention, offering interpretable sparse attention and competitive performance on graph benchmarks.

This paper proposes DSiGAT, a dynamic scene graph attention framework that jointly predicts lane-change intentions and future trajectories for all interacting vehicles in a traffic scene, achieving state-of-the-art results on NGSIM and highD datasets.