Edge-aware Decoding for Neural Asymmetric Routing
Researchers have developed an edge-aware decoder to improve neural asymmetric routing models. This new decoder explicitly incorporates transition-level information, such as the current directed edge and closure cues, into the final routing decision. When tested on the Asymmetric Traveling Salesperson Problem (ATSP), this approach reduced the performance gap significantly, demonstrating the importance of decision-time exposure to edge information. AI
IMPACT Improves performance on complex routing problems, potentially impacting logistics and optimization AI.