Título:
LASE : Learned Adjacency Spectral Embeddings
Otros títulos:
Coordinador:
Director:
Compilador:
Autor:
Pérez Casulo, María Sofía
Fiori, Marcelo
Larroca, Federico
Mateos, Gonzalo
Fiori, Marcelo
Larroca, Federico
Mateos, Gonzalo
Tutor:
Tipo de documento:
Artículo
Editor:
Palabras clave:
Graph Representation Learning
Algorithm Unrolling
Gradient Descent
Algorithm Unrolling
Gradient Descent
Descriptores:
Año de publicación:
2025
Contenido:
Resumen:
We put forth a principled design of a neural architecture to learn nodal Adjacency Spectral Embeddings (ASE) from graph inputs. By bringing to bear the gradient descent (GD) method and leveraging the technique of algorithm unrolling, we truncate and re-interpret each GD iteration as a layer in a graph neural network (GNN) that is trained to approximate the ASE. Accordingly, we call the resulting embeddings and our parametric model Learned ASE (LASE), which is interpretable, parameter efficient, robust to inputs with unobserved edges, and offers controllable complexity during inference. LASE layers combine Graph Convolutional Network (GCN) and fully-connected Graph Attention Network (GAT) modules, which is intuitively pleasing since GCN-based local aggregations alone are insufficient to express the sought graph eigenvectors. We propose several refinements to the unrolled LASE architecture (such as sparse attention in the GAT module and decoupled layerwise parameters) that offer favorable approximation error versus computation tradeoffs; even outperforming heavily-optimized eigendecomposition routines from scientific computing libraries. Because LASE is a differentiable function with respect to its parameters as well as its graph input, we can seamlessly integrate it as a trainable module within a larger (semi-)supervised graph representation learning pipeline. The resulting end-to-end system effectively learns "discriminative ASEs" that exhibit competitive performance in supervised link prediction and node classification tasks, outperforming a GNN even when the latter is endowed with open loop, meaning task-agnostic, precomputed spectral positional encodings.
Descripción:
metadata.articulos.dc.description.uri:
Editorial:
OpenReview
EN:
Transactions on Machine Learning Research, jun. 2025, pp. 1-31.
Financiadores:
CSIC (I+D proyecto 22520220100076UD)
Citación:
Pérez Casulo, M., Fiori, M., Larroca, F. y otros. "LASE : Learned Adjacency Spectral Embeddings". Transactions on Machine Learning Research. [en línea]. 2025, pp. 1-31.
Citación:
metadata.articulos.cc.license.name:
ISBN:
e-ISBN:
ISSN:
2835-8856
ISMN:
Otros identificadores:
Cobertura geográfica:
Cobertura temporal:
Departamento académico:
Telecomunicaciones
Grupo de investigación:
Análisis de Redes, Tráficos y Estadísticas de Servicios (ARTES)
Licencia:
Licencia Creative Commons Atribución (CC - By 4.0)
Colecciones:
| Ficheros | Descripción | Tamaño | Formato | ||
|---|---|---|---|---|---|
| PFLM25.pdf | Camera-Ready | 4.64 MB | Adobe PDF |