Neighbour-Driven Gaussian Process Variational Autoencoders for Scalable Structured Latent Modelling

2025-05-22

AI Summary

To address the computational intractability of exact Gaussian process (GP) inference in large-scale GP variational autoencoders (GPVAEs), this paper proposes a neighborhood-driven local approximation method: for each data point, GP inference is performed only over its *k*-nearest neighbors in the latent space—bypassing global kernel assumptions and eliminating the need for large sets of inducing points. This is the first work to incorporate neighborhood locality into the GPVAE variational inference framework, enabling more flexible kernel design while preserving essential latent variable dependencies. Experiments demonstrate substantial improvements over existing GPVAE baselines across representation learning, missing data imputation, and conditional generation tasks: prediction accuracy increases significantly, training speed accelerates by 3–5×, and memory consumption decreases by approximately 40%.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Gaussian Process (GP) Variational Autoencoders (VAEs) extend standard VAEs by replacing the fully factorised Gaussian prior with a GP prior, thereby capturing richer correlations among latent variables. However, performing exact GP inference in large-scale GPVAEs is computationally prohibitive, often forcing existing approaches to rely on restrictive kernel assumptions or large sets of inducing points. In this work, we propose a neighbour-driven approximation strategy that exploits local adjacencies in the latent space to achieve scalable GPVAE inference. By confining computations to the nearest neighbours of each data point, our method preserves essential latent dependencies, allowing more flexible kernel choices and mitigating the need for numerous inducing points. Through extensive experiments on tasks including representation learning, data imputation, and conditional generation, we demonstrate that our approach outperforms other GPVAE variants in both predictive performance and computational efficiency.

Problem

• Scalable inference in Gaussian Process VAEs
• Reducing computational cost of GPVAE models
• Preserving latent dependencies with neighbor-driven approximation

Innovation

• Neighbour-driven approximation for scalable GPVAE inference
• Local adjacencies preserve latent dependencies
• Flexible kernel choices reduce inducing points need