![[2511.06781] On the Mechanisms of Collaborative Learning in VAE Recommenders](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2511.06781] On the Mechanisms of Collaborative Learning in VAE Recommenders
Summary
This paper explores the mechanisms of collaborative learning in Variational Autoencoder (VAE) recommenders, highlighting the role of latent proximity and proposing methods to enhance user collaboration.
Why It Matters
Understanding collaborative learning in VAE recommenders is crucial for improving recommendation systems. This research provides insights into optimizing user interactions and enhancing algorithm performance, which can lead to better user experiences in platforms like Netflix and Amazon.
Key Takeaways
- Collaboration in VAE recommenders is influenced by latent proximity among users.
- The study introduces an anchor regularizer to stabilize user identities while enhancing global consistency.
- Two mechanisms for encouraging global mixing are analyzed, each with distinct trade-offs.
Computer Science > Machine Learning arXiv:2511.06781 (cs) [Submitted on 10 Nov 2025 (v1), last revised 14 Feb 2026 (this version, v2)] Title:On the Mechanisms of Collaborative Learning in VAE Recommenders Authors:Tung-Long Vuong, Julien Monteil, Hien Dang, Volodymyr Vaskovych, Trung Le, Vu Nguyen View a PDF of the paper titled On the Mechanisms of Collaborative Learning in VAE Recommenders, by Tung-Long Vuong and 5 other authors View PDF HTML (experimental) Abstract:Variational Autoencoders (VAEs) are a powerful alternative to matrix factorization for recommendation. A common technique in VAE-based collaborative filtering (CF) consists in applying binary input masking to user interaction vectors, which improves performance but remains underexplored theoretically. In this work, we analyze how collaboration arises in VAE-based CF and show it is governed by \emph{latent proximity}: we derive a latent sharing radius that informs when an SGD update on one user strictly reduces the loss on another user, with influence decaying as the latent Wasserstein distance increases. We further study the induced geometry: with clean inputs, VAE-based CF primarily exploits \emph{local} collaboration between input-similar users and under-utilizes \emph{global} collaboration between far-but-related users. We compare two mechanisms that encourage \emph{global} mixing and characterize their trade-offs: \ding{172} $\beta$-KL regularization directly tightens the information bottleneck, promoting p...
