α^α-Rank: Practically Scaling α-Rank through Stochastic Optimisation

09/25/2019
by   Yaodong Yang, et al.
0

Recently, α-Rank, a graph-based algorithm, has been proposed as a solution to ranking joint policy profiles in large scale multi-agent systems. α-Rank claimed tractability through a polynomial time implementation with respect to the total number of pure strategy profiles. Here, we note that inputs to the algorithm were not clearly specified in the original presentation; as such, we deem complexity claims as not grounded, and conjecture solving α-Rank is NP-hard. The authors of α-Rank suggested that the input to α-Rank can be an exponentially-sized payoff matrix; a claim promised to be clarified in subsequent manuscripts. Even though α-Rank exhibits a polynomial-time solution with respect to such an input, we further reflect additional critical problems. We demonstrate that due to the need of constructing an exponentially large Markov chain, α-Rank is infeasible beyond a small finite number of agents. We ground these claims by adopting amount of dollars spent as a non-refutable evaluation metric. Realising such scalability issue, we present a stochastic implementation of α-Rank with a double oracle mechanism allowing for reductions in joint strategy spaces. Our method, α^α-Rank, does not need to save exponentially-large transition matrix, and can terminate early under required precision. Although theoretically our method exhibits similar worst-case complexity guarantees compared to α-Rank, it allows us, for the first time, to practically conduct large-scale multi-agent evaluations. On 10^4×10^4 random matrices, we achieve 1000x speed reduction. Furthermore, we also show successful results on large joint strategy profiles with a maximum size in the order of O(2^25) (33 million joint strategies) – a setting not evaluable using α-Rank with reasonable computational budget.

READ FULL TEXT

page 1

page 14

research
09/25/2019

α^α-Rank: Scalable Multi-agent Evaluation through Evolution

Although challenging, strategy profile evaluation in large connected lea...
research
03/04/2019

α-Rank: Multi-Agent Evaluation by Evolution

We introduce α-Rank, a principled evolutionary dynamics methodology, for...
research
04/09/2021

Ranking Bracelets in Polynomial Time

The main result of the paper is the first polynomial-time algorithm for ...
research
05/25/2023

Corrigendum to "On the monophonic rank of a graph" [Discrete Math. Theor. Comput. Sci. 24:1 (2022) #3]

In this corrigendum, we give a counterexample to Theorem 5.2 in “On the ...
research
09/26/2022

Maker-Breaker is solved in polynomial time on hypergraphs of rank 3

In the Maker-Breaker positional game, Maker and Breaker take turns picki...
research
02/13/2017

Certificates for triangular equivalence and rank profiles

In this paper, we give novel certificates for triangular equivalence and...

Please sign up or login with your details

Forgot password? Click here to reset