In many advanced recommender systems (e.g., GNN-based), local smoothing mechanisms would distill collaborative signals but could inadvertently amplify targeted data poisoning threats. Existing defenses predominantly relied on rigid binary filtration strategies. However, such structural deletion of suspicious nodes often compromises graph connectivity, leading to a severe robustness-fidelity dilemma. To fill this gap, we propose SiPo (Silencing the Poison), a novel framework that shifts the paradigm from node removal to gradient silencing. SiPo leverages unsupervised granular ball computing to adaptively capture adversarial topological micro-clusters within the feature space. Subsequently, it geometrically collapses high-risk nodes towards a non-informative singularity via a global gravity gating mechanism. This operation blocks the backpropagation path of malicious gradients, effectively suppressing attacks while preserving the collaborative signals of normal users. Theoretical analysis confirms that SiPo enforces a compression of the local Lipschitz constant. Furthermore, extensive experiments on four public datasets demonstrate that SiPo not only significantly reduces attack success rates but also consistently improves recommendation performance. The implementation code is available at: https://github.com/KREML0/SiPo.
Yunmeng Zhao, Jun Shen, Shenbao Yu et al.· Proceedings of the 32nd ACM...· 0 citations
Spatial domain identification in spatial transcriptomics (ST) partitions a tissue section into spatially coherent regions with distinct transcriptional programs. Recent graph-based approaches construct a spatial neighborhood graph and/or an expression-similarity graph and perform unsupervised domain identification with GNN encoders or graph-regularized embeddings. However, dropout and boundary mixing make neighborhood reliability highly non-uniform across space, so fixed graphs can propagate signals across true boundaries and blur domains, motivating unit-wise multi-view fusion and topology refinement. We propose GatorTrio, a topology-refined tri-view graph learning framework that adaptively fuses views per unit via an interaction-aware sparse Mixture-of-Experts (MoE) router. GatorTrio constructs three complementary graphs: (i) an attention-induced expression-affinity graph, (ii) a dropout-aware expression-similarity graph, and (iii) a spatial kNN neighborhood graph. A parameter-sharing GNN encoder encodes each graph, and the router fuses the resulting view-specific embeddings; view dropout and learnable missing-view tokens improve robustness to noisy and missing views. We refine topology through self-training by pruning pseudo-label–inconsistent inter-cluster edges and re-encoding on the refined graphs. Training combines bidirectional topology-informed contrastive learning between expression views with spatially anchored prototype alignment that matches expression-view centroids to perturbed spatial prototypes under a curriculum warm-up. Experiments on diverse ST benchmarks show consistent improvements over ten representative baselines and support downstream biological analyses.
Zhenhao Zhang, Jun Shen, Jiang Bian et al.· Proceedings of the 32nd ACM...· 0 citations