EMS: Multi-Agent Voting via Efficient Majority-then-Stopping
Researchers have developed a new multi-agent voting system called Efficient Majority-then-Stopping (EMS) to reduce computational overhead. EMS improves reasoning efficiency by first ordering agents based on their historical reliability for similar queries and then invoking them in that order. The system terminates voting once a leading answer cannot be surpassed, returning the consensus decision. Evaluations show EMS maintains accuracy while significantly reducing the number of agents invoked and token consumption. AI
IMPACT Reduces computational costs for multi-agent systems, potentially enabling more complex and efficient AI applications.