---
title: "PRIME: Plasticity Recovery in Multi-Agent Environments for UAV-Assisted Emergency Communication Networks"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.17922v1"
permalink: "https://feed.astralor.com/items/itm_b7c6e8d7d7044153246ef00b"
published: "2026-07-20T13:19:58.000Z"
score: 76
channel: "tech"
tags: ["PRIME","Plasticity Recovery","Multi-Agent Reinforcement Learning","UAV Communication","Silent Neuron"]
---

# PRIME: Plasticity Recovery in Multi-Agent Environments for UAV-Assisted Emergency Communication Networks

## AI commentary

> This paper's PRIME method cleverly leverages gradient signals (not hand-crafted proxies) to judge neuron activity, avoiding training instability from brute-force resets in multi-agent settings. For researchers in distributed AI systems and robot scheduling, it advances the intersection of continual learning and multi-agent RL, with a theoretical bound that ensures scalability—definitely worth a read.

## At a glance

Most reinforcement learning controllers for UAV-assisted emergency communication networks assume stationary conditions, and the few that handle change only react to the external environment while ignoring the network's internal state. This paper shows that sustained non-stationarity damages the internal state directly: as objectives shift, neurons progressively fall dormant and the shared policy loses learning capacity. Simply resetting dormant neurons is unsafe under shared-parameter multi-agent training, as many seemingly inactive neurons still receive strong gradients, and dormancy depends on observations. PRIME extends the bidirectional Silent Neuron framework to cooperative multi-agent RL, aggregating activation and gradient statistics over the full team batch, and reinitializes only neurons that are simultaneously activation-dormant and gradient-silent. On a phase-switching UAV emergency communication simulator, PRIME improves interquartile mean return by 24.9% over MAPPO and reduces dormant neuron fractions from 40-45% to 10-20%. A dynamic regret bound shows perturbation cost scales with the small silent-subspace dimension rather than full parameter count.

## Key points

- PRIME selectively reinitializes neurons using a bidirectional criterion (activation-dormant and gradient-silent), avoiding safety issues from brute-force resetting.
- On a UAV emergency communication simulator, PRIME reduces dormant neuron fraction from 40-45% to 10-20% and improves interquartile mean return by 24.9% over MAPPO.
- A dynamic regret bound shows perturbation cost scales with the small silent-subspace dimension, not the full parameter count, providing theoretical scalability.

## Full text

Most reinforcement learning controllers for these networks assume stationary conditions, and the few that handle change react to the external environment while leaving the network's internal state unexamined. We show that sustained non-stationarity damages this internal state directly: as objectives shift, neurons progressively fall dormant and the shared policy loses the capacity to learn. The obvious remedy, resetting dormant neurons, is unsafe under shared-parameter multi-agent training: many neurons that appear inactive are still receiving strong training gradients, and whether a neuron appears dormant depends on which agent's observations it processes. PRIME (Plasticity Recovery In Multi-agent Environments) therefore verifies both directions before intervening. Extending the bidirectional Silent Neuron framework to cooperative multi-agent reinforcement learning, it aggregates activation and gradient statistics over the full team batch, reads the backward signal from the gradient the training loss has already deposited , not from a hand-crafted proxy, and reinitializes only neurons that are simultaneously activation-dormant and gradient-silent. Useful representations are preserved while learning capacity is restored. On a phase-switching UAV emergency communication simulator, PRIME improves interquartile mean return by 24.9\% over MAPPO and holds dormant neuron fractions at 10--20\% versus 40--45\%; ablations attribute the gains to the gradient signal and team-level aggregation rather than to the specific reset operator. A dynamic regret bound shows that the perturbation cost scales with the small silent-subspace dimension rather than the full parameter count.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_b7c6e8d7d7044153246ef00b