.png){kind=icon}
.png){kind=icon}
Return
Federated Neural Nonparametric Point Processes
.png){kind=icon}
OA
AI
DOI：10.1016/j.artint.2025.104454.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We introduce FedPP, the first federated adaptation of Temporal Point Process (TPP) models, which bridges the gap between TPPs and Federated Learning while addressing key challenges such as event sparsity, uncertainty, and privacy concerns. • We propose an innovative integration of neural embedding techniques within the kernels of sigmoidal Gaussian Cox processes (SGCPs), which significantly enhances their expressiveness and enables the effective utilization of historical data. • We develop a divergence-guided global aggregation mechanism, to facilitate the secure sharing of neural embedding distributions between the server and clients, which ensures robust global modeling while preserving client-specific privacy. • Our method outperforms existing approaches on benchmark datasets, capturing event sparsity and uncertainty in federated environments without compromising privacy.
Keywords:
Federated learning
Temporal point processes
Federated neural point processes
Neural embedding
Gaussian processes
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
A
IF:
4.6
Papers:
75
Citations:
1
Organization
View more organizations