The robustness of Internet of Things (IoT) communication topologies against internal failures and external perturbations is a fundamental prerequisite for maintaining system stability. This paper studies IoT topology robustness from two perspectives: robustness metric and optimization. Existing robustness metrics, primarily based on the maximum connected subgraph, neglect contributions from other connected subgraphs, thereby inadequately capturing dynamic topological changes. To address this issue, we propose a robustness metric based on topology data reachability, which sensitively reflects the data transmission capability of an IoT topology under arbitrary perturbations. Regarding robustness optimization, most existing methods adopt centralized strategies that rely on global information, resulting in inefficiencies and limited adaptability in decentralized IoT environments. We propose DecTRO, a Decentralized Topology Robustness Optimization method for IoT via multi-agent graph reinforcement learning. To mitigate partial observability, DecTRO employs a scalable graph attention network enhanced with multi-modal sampling, which aggregates cross-agent information and captures spatiotemporal correlations. Furthermore, a topology robustness-oriented node sampling method is introduced to reduce action-space complexity and accelerate convergence, while a decentralized heuristic reward function enables efficient online decentralized learning. Experimental results show that DecTRO achieves up to two orders of magnitude (5–125×) greater improvement in robustness per unit time compared with state-of-the-art baselines, striking a favorable balance between robustness enhancement and computational efficiency.

Backdoor attacks present significant risks to the security of deep neural networks (DNNs) in NLP domain, as the attackers can covertly manipulate the model

Graph neural networks (GNNs) are highly effective on relational data, yet their performance degrades sharply when graph topology is poisoned before training. Existing defenses usually assume a fixed attack pattern and a fixed graph structure, which makes them brittle when the poisoned graph changes across attacks, perturbation budgets, or deployment conditions. We propose GADD, a game-inspired adversarial distillation framework for robust graph defense. GADD first constructs multiple positive and negative graph views through a homophily-aware graph sampling scheme, allowing the model to learn from both purified and high-risk subgraphs. It then trains a heterogeneous group of student GNNs online, where each student receives global class-distribution knowledge from its peers and local structural knowledge through an adversarial cyclic distillation objective. Finally, GADD replaces uniform ensembling with an entropy-regularized adaptive aggregation rule that assigns graph-adaptive weights according to confidence and inter-model agreement. On Cora, CiteSeer, and PubMed, GADD consistently improves robustness against both Meta and Nettack attacks while preserving clean accuracy. Under the strongest Meta and Nettack settings in the main benchmark, GADD improves the best competing baseline by up to 2.99 and 3.42 percentage points, respectively. Additional ablations show that graph sampling, adversarial distillation, and adaptive aggregation all contribute materially to the final robustness gains.