{"total":12,"items":[{"citing_arxiv_id":"2605.21435","ref_index":74,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"Gaussian Sheaf Neural Networks","primary_cat":"cs.LG","submitted_at":"2026-05-20T17:26:32+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Gaussian Sheaf Neural Networks derive a sheaf Laplacian for Gaussian node features on graphs to preserve their geometric structure during message passing.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.19822","ref_index":34,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"ST-TGExplainer: Disentangling Stability and Transition Patterns for Temporal GNN Interpretability","primary_cat":"cs.LG","submitted_at":"2026-05-19T13:16:22+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ST-TGExplainer disentangles stability and transition patterns in temporal graphs via a self-explainable TGNN guided by a disentangled information bottleneck objective to produce more faithful explanations.","context_count":1,"top_context_role":"other","top_context_polarity":"unclear","context_text":"Consider the KL divergence between pθ(GE|Gt)andq(G E): Ep(Gt) h DKL pθ(GE|Gt)∥q(G E) \u0001i =E p(Gt)Epθ(GE |Gt) \u0014 log pθ(GE|Gt) q(GE) \u0015 =E p(Gt,GE ) \u0014 log pθ(GE|Gt) q(GE) \u0015 =E p(Gt,GE ) \u0014 log pθ(GE|Gt) pθ(GE) \u0015 +E p(Gt,GE ) \u0014 log pθ(GE) q(GE) \u0015 =E p(Gt,GE ) \u0014 log pθ(GE|Gt) pθ(GE) \u0015 +E p(GE ) \u0014 log pθ(GE) q(GE) \u0015 =I(G t;G E) +D KL pθ(GE)∥q(G E) \u0001 ≥I(G t;G E), (34) where the inequality follows from the non-negativity of KL divergence:D KL pθ(GE)∥q(G E) \u0001 ≥0. Step 4: Combining the bounds.Combining Eq. (32) and Eq. (34), we obtain the chain of inequalities: I(G t;G S,G T )≤I(G t;G E)≤E p(Gt) h DKL pθ(GE|Gt)∥q(G E) \u0001i .(35) Therefore, minimizing the compression loss: LCom :=E p(Gt) h DKL pθ(GE|Gt)∥q(G E) \u0001i"},{"citing_arxiv_id":"2605.18893","ref_index":25,"ref_count":2,"confidence":0.35,"is_internal_anchor":false,"paper_title":"Position: Graph Condensation Needs a Reset -- Move Beyond Full-dataset Training and Model-Dependence","primary_cat":"cs.LG","submitted_at":"2026-05-17T07:08:22+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"The paper claims current graph condensation approaches are flawed due to full-dataset training requirements, high overhead, poor generalization, and misleading evaluation metrics, calling for a reset toward lightweight and architecture-agnostic methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.14594","ref_index":78,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"TOPOS: High-Fidelity and Efficient Industry-Grade 3D Head Generation","primary_cat":"cs.CV","submitted_at":"2026-05-14T09:02:32+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"TOPOS creates high-fidelity 3D heads with fixed industry topology from single images via a specialized VAE with Perceiver Resampler and a rectified flow transformer.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12061","ref_index":109,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory","primary_cat":"cs.AI","submitted_at":"2026-05-12T12:47:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"SAGE is a self-evolving agentic graph-memory engine that dynamically constructs and refines structured memory graphs via writer-reader feedback, yielding performance gains on multi-hop QA, open-domain retrieval, and long-term agent benchmarks.","context_count":1,"top_context_role":"background","top_context_polarity":"unclear","context_text":"Taking the maximum overvgives the conclusion. 29 E.4 Single-layer Stability of Structurally Gated Propagation Lemma E.9(Boundedness and stability of structural gate).The structural gate of layerl, g(l) uv = 1 +δtanh(MLP (l) g (z(l) uv)),(107) satisfies g(l) uv ∞ ≤1 +δ,(108) and g(l) uv −g ′(l) uv ∞ ≤δL g,l z(l) uv −z ′(l) uv 2 .(109) Proof. Since the range of tanh is contained in [−1,1] , the first statement follows immediately. Moreover, because tanh is 1-Lipschitz and MLP(l) g is Lg,l-Lipschitz in the trajectory neighborhood, g(l) uv −g ′(l) uv ∞ ≤δ MLP(l) g (z(l) uv)−MLP (l) g (z′(l) uv ) ∞ ≤δL g,l z(l) uv −z ′(l) uv 2 .(110) Lemma E.10(Single-layer stability of structurally gated propagation)."},{"citing_arxiv_id":"2605.10247","ref_index":34,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"Teaching LLMs to See Graphs: Unifying Text and Structural Reasoning","primary_cat":"cs.LG","submitted_at":"2026-05-11T09:19:55+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"GTLM injects graph-aware attention biases into LLMs using only 0.015% extra parameters, enabling native graph processing that matches 7B models with a 1B model on text-attributed graph benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.10095","ref_index":69,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"Learning to Compress and Transmit: Adaptive Rate Control for Semantic Communications over LEO Satellite-to-Ground Links","primary_cat":"cs.NI","submitted_at":"2026-05-11T07:11:11+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"RL agent adaptively controls compression rate in semantic satellite communications to achieve 95% qualified image frames with no packet loss by using SNR predictions and queue management.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.09907","ref_index":24,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"RADAR: Redundancy-Aware Diffusion for Multi-Agent Communication Structure Generation","primary_cat":"cs.AI","submitted_at":"2026-05-11T02:50:40+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.08679","ref_index":16,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"Attention-based graph neural networks: a survey","primary_cat":"cs.SI","submitted_at":"2026-05-09T04:33:14+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"The survey groups attention-based GNNs into three stages—graph recurrent attention networks, graph attention networks, and graph transformers—while reviewing architectures and future directions.","context_count":1,"top_context_role":"method","top_context_polarity":"use_method","context_text":"In addition, the distribution function con- verts the attention score to the attention coefficient, which makes the attention coefficients comparable across different nodes. Finally, the weighted sum func- tion is used to update and aggregate the representations of nodes in the graph. The above local attention layer can be defined as follows: hl i = Θl xl i , h l j = Θl xl j (16) scoresl ij =Sim hl i, hl j \u0001 (17) αl ij =N orm scoresl\u0001 (18) hl+1 i =σ    X j∈Γi αl ij hl j    (19) whereσdenotes the non-linear activation function. Γ i refers to the local neigh- borhoods of the given nodev i, which is exactly the first-order node in GAT. Sim(·) represents the alignment functions (listed in Table 1) andN orm(·) is the distribution functions, such as softmax, and sigmoid."},{"citing_arxiv_id":"2605.05476","ref_index":50,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"A Unified Benchmark for Evaluating Knowledge Graph Construction Methods and Graph Neural Networks","primary_cat":"cs.LG","submitted_at":"2026-05-06T21:53:49+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A dual-purpose benchmark supplies two text-derived knowledge graphs and one expert reference graph on the same biomedical corpus to jointly measure construction method quality and GNN robustness via semi-supervised node classification.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.02617","ref_index":19,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"SCGNN: Semantic Consistency enhanced Graph Neural Network Guided by Granular-ball Computing","primary_cat":"cs.AI","submitted_at":"2026-05-04T14:09:56+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"SCGNN uses granular-ball computing to partition nodes into groups, builds an anchor-based augmented graph, and fuses predictions with label-consistency supervision to improve semantic consistency in GNNs.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.17715","ref_index":53,"ref_count":1,"confidence":0.35,"is_internal_anchor":false,"paper_title":"Program Structure-aware Language Models: Targeted Software Testing beyond Textual Semantics","primary_cat":"cs.SE","submitted_at":"2026-04-20T01:54:01+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"GLMTest integrates code property graphs and GNNs with LLMs to steer test case generation toward targeted branches, raising branch accuracy from 27.4% to 50.2% on the TestGenEval benchmark.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}