Skip to content
ml-recall

#credit-assignment

3 questions from Multi-Agent Systems, RL Fundamentals

Timed test

3 questions · 3 min · score and explanations at the end

Take a test
Q 1 / 3Score 0 / 0

Question 1: Short answer

hardAI-checkedMulti-Agent Systems

In a cooperative task all agents receive one shared team reward, and COMA gives each agent its own advantage through a counterfactual baseline. Agent aa has three actions. With the other agents' actions u−a\mathbf{u}^{-a} held fixed, the centralised critic and agent aa's policy give:

agent aa's action u′au'^a u1u_1 u2u_2 u3u_3
Q(s,(u−a,u′a))Q(s, (\mathbf{u}^{-a}, u'^a)) 4.0 6.0 1.0
πa(u′a∣τa)\pi^a(u'^a \mid \tau^a) 0.2 0.5 0.3

Agent aa actually took u2u_2. What is its COMA advantage?

a number, 2 decimals