CoLFA
A lightweight fault attribution method for multi-agent systems. It organizes input using collaboration cues such as handoffs and cross-references to precisely locate erroneous steps and responsible agents at low token cost.
Timeline (last 90 days)
CoLFA is a lightweight multi-agent fault attribution method guided by collaborative cues, which organizes attribution inputs by leveraging collaborative cues.
CoLFA achieves lightweight attribution through two key techniques: Handoff-Guided Step Expansion and Cross-Referenced Content Organization.
On the two datasets Who&When and LongRCA, across five settings and three backbone models, CoLFA achieved the highest exact step localization accuracy while consuming the fewest tokens.
Compared with the individual baselines under each setting, CoLFA improved exact step localization accuracy by up to 37.50%.
Compared with the individual baselines under each setting, CoLFA reduced token consumption by up to 97.6%.
All Facts (5)
CoLFA is a lightweight multi-agent fault attribution method guided by collaborative cues, which organizes attribution inputs by leveraging collaborative cues.
50%UnverifiedCoLFA achieves lightweight attribution through two key techniques: Handoff-Guided Step Expansion and Cross-Referenced Content Organization.
50%UnverifiedOn the two datasets Who&When and LongRCA, across five settings and three backbone models, CoLFA achieved the highest exact step localization accuracy while consuming the fewest tokens.
50%UnverifiedCompared with the individual baselines under each setting, CoLFA improved exact step localization accuracy by up to 37.50%.
50%UnverifiedCompared with the individual baselines under each setting, CoLFA reduced token consumption by up to 97.6%.
50%