A graph-native generative compression framework that conditions a graph denoising diffusion probabilistic model (DDPM) decoder on latent representations of continuous graph-structured signals and shows that the resulting encoded representations are higher-fidelity and more distributionally aligned than those of vanilla autoencoders.
It is shown that using the standard velocity prediction in flow matching in this high-dimensional space requires the model to fit orthogonal noise directions outside the low-dimensional signal manifold, making optimization inefficient, and motivated using the clean data parameterization ($\boldsymbol{x}_{0}$-prediction...
Chao Feng, Zhi-Yang Xu, Bo-Wei Chen et al.· 0 citations
The results show that compression and execution layout must be co-designed to convert model-size reduction into practical parallel inference acceleration, and that compression and execution layout must be co-designed to convert model-size reduction into practical parallel inference acceleration.
Qian-Sheng Song, Guo-Lin Tang· Proceedings of the Internati...· 0 citations
Across five graph benchmarks, the performance of SDRs is competitive with that of real-valued dense embeddings on node classification and link prediction, while the architecture exhibits portability, resilience to noise, and robustness to data corruption.
This work proves three core properties of HouseGNN: every internal layer preserves the node-wise Euclidean norm; the Householder reflector is scale scale and sign-invariant in the message; and pairwise distance between nodes can change through mismatch between node-wise orthogonal operators.
Experiments show that incorporating NC-LID-based regularization consistently improves reconstruction performance over the baseline without structural regularization and the method using hub-aware regularization, which highlights NC-LID as a useful structural signal for enhancing distance-based graph autoencoders in dyn...
Aleksandar Tomčić, Milos Savic, Milos Radovanovic· 0 citations
This study proposes a model called the Residual compensation graph Convolutional Generalized Extreme Learning Machine Autoencoder (RCGELM-AE), which is built on the Extreme Learning Machine Autoencoder (ELM-AE) and incorporates two key designs, and offers a novel, gradient-descent-free solution for achieving efficient...
