Tag
The paper provides a completeness theory proving that multi-layer message passing in graph neural networks achieves universal approximation for interatomic potentials, justifying common architectural practices in machine-learned potentials.
The paper introduces Riemannian Hodge Message Passing (RHMP), a neural architecture that separates topology and geometry for learning physical fields on meshes, achieving superior performance across multiple benchmarks.
This theoretical paper proves that ReLU-based message-passing GNNs are strictly more expressive than GNNs using any eventually constant activation functions (e.g., truncated ReLU) with respect to Boolean queries, even on Boolean-featured graphs.
This paper introduces addressable and cardinality-preserving global memory for message-passing neural networks via cross-attention slots, addressing the finite-capacity bottleneck of virtual nodes and improving performance on multiplicity-aware tasks.
This paper proves that graph neural networks with random node features can universally approximate permutation-invariant or equivariant functions on directed graphs, and provides approximation rate bounds for differentiable functions.
This paper studies the efficacy of various Graph Neural Network message-passing layers in regression contexts, finding that deep convolutional GNNs, particularly GEN, outperform attention-based GNNs.
HypNO introduces a graph-based neural operator that uses physics-informed message passing on a space-time finite-volume cell graph to solve scalar hyperbolic conservation laws, accurately capturing shocks and discontinuities. The method is benchmarked on LWR and ARZ traffic-flow models.
Proposes a message-passing-based two-timescale Bayesian deep learning framework for joint channel and memory hardware impairment tracking in massive MIMO systems.
GRAB uses a GNN encoder to convert relational tables into latent tokens for frozen LLMs, achieving significant performance gains in multi-table question answering.
This paper proposes a K-Hop Gaussian (KHG) diffusion kernel as a preprocessing module for graph neural networks, balancing local and global information propagation to mitigate over-smoothing and information bottlenecks. Experiments show significant improvements over traditional message-passing GNNs and existing diffusion kernels, especially on noisy or structurally complex graphs.
This paper analyzes Active Inference by proving that the Variational Free Energy of an augmented generative model can be decomposed into the predictive model's VFE plus explicit entropy-correction terms, yielding a full variational characterization of Expected Free Energy-based planning. The authors derive a message-passing scheme for EFE-based planning and validate it on grid-world environments.
Proposes a node-level spectral energy formulation for detecting camouflaged anomalies in graphs, extending to spatio-temporal settings with energy-driven message passing. Demonstrates effectiveness on large-scale benchmarks.
The paper introduces a novel task of fact generation for hyper-relational knowledge graphs (HKGs) and proposes KREPE, a generative representation learning method using masked discrete diffusion that unifies link prediction and fact generation, achieving state-of-the-art performance.