Graph-SND: Sparse Aggregation for Behavioral Diversity in Multi-Agent Reinforcement Learning
Shawn Ray · 2026
Read Graph-SND: Sparse Aggregation for Behavioral Diversity in Multi-Agent Reinforcement Learning by Shawn Ray (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-05020.txt.
Description
Shawn Ray makes a diversity measure for multi-agent reinforcement learning cheaper to compute by replacing a complete comparison graph with a sparser one. The standard measure averages distances between every pair of agents, so its cost grows quadratically with team size. Graph-SND instead averages over the edges of any chosen graph, weighted as set.
Using the full graph recovers the original measure exactly, while a fixed sparse graph gives a local version at edge-count cost. Random edge samples give unbiased estimates with stated concentration bounds, and Ray proves distortion results for expander graphs. In large team training runs, sampled Graph-SND tracked the full measure at about a tenth of the time.
Catalog record
| CREATOR | Shawn Ray |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.05020