# On the Mechanisms of Collaborative Learning in VAE Recommenders

arXiv: https://arxiv.org/abs/2511.06781 · HTML: https://mlsanity.com/paper/2511.06781/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 51 references that are in the map.
- First posted on arXiv 10 Nov 2025.

## Builds on (most cited first, 20 of 20)

- [Variational Inference with Normalizing Flows](https://mlsanity.com/paper/1505.05770/) (2015)
- [Neural Collaborative Filtering](https://mlsanity.com/paper/1708.05031/) (2017)
- [Improving Variational Inference with Inverse Autoregressive Flow](https://mlsanity.com/paper/1606.04934/) (2016)
- [Variational Autoencoders for Collaborative Filtering](https://mlsanity.com/paper/1802.05814/) (2018)
- [VAE with a VampPrior](https://mlsanity.com/paper/1705.07120/) (2017)
- [Learning Disentangled Representations for Recommendation](https://mlsanity.com/paper/1910.14238/) (2019)
- [Embarrassingly Shallow Autoencoders for Sparse Data](https://mlsanity.com/paper/1905.03375/) (2019)
- [RecVAE: a New Variational Autoencoder for Top-N Recommendations with Implicit Feedback](https://mlsanity.com/paper/1912.11160/) (2019)
- [The Usual Suspects? Reassessing Blame for VAE Posterior Collapse](https://mlsanity.com/paper/1912.10702/) (2019)
- [Collaborative Retrieval for Large Language Model-based Conversational Recommender Systems](https://mlsanity.com/paper/2502.14137/) (2025)
- [Alleviating Cold-start Problem in CTR Prediction with A Variational Embedding Learning Framework](https://mlsanity.com/paper/2201.10980/) (2022)
- [Enhancing VAEs for Collaborative Filtering: Flexible Priors & Gating Mechanisms](https://mlsanity.com/paper/1911.00936/) (2019)
- [Causal Disentangled Variational Auto-Encoder for Preference Understanding in Recommendation](https://mlsanity.com/paper/2304.07922/) (2023)
- [Item Recommendation with Variational Autoencoders and Heterogenous Priors](https://mlsanity.com/paper/1807.06651/) (2018)
- [On Posterior Collapse and Encoder Feature Dispersion in Sequence VAEs](https://mlsanity.com/paper/1911.03976/) (2019)
- [DualVAE: Dual Disentangled Variational AutoEncoder for Recommendation](https://mlsanity.com/paper/2401.04914/) (2024)
- [Towards Amortized Ranking-Critical Training for Collaborative Filtering](https://mlsanity.com/paper/1906.04281/) (2019)
- [Deep Variational Autoencoder with Shallow Parallel Path for Top-N Recommendation (VASP)](https://mlsanity.com/paper/2102.05774/) (2021)
- [MARec: Metadata Alignment for cold-start Recommendation](https://mlsanity.com/paper/2404.13298/) (2024)
- [Geometric Collaborative Filtering with Convergence](https://mlsanity.com/paper/2410.03064/) (2024)

## Abstract (arXiv)

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 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 local collaboration between input-similar users and under-utilizes global collaboration between far-but-related users. We compare two mechanisms that encourage global mixing and characterize their trade-offs: ding172 β-KL regularization directly tightens the information bottleneck, promoting posterior overlap but risking representational collapse if too large; ding173 input masking induces stochastic geometric contractions and expansions, which can bring distant users onto the same latent neighborhood but also introduce neighborhood drift. To preserve user identity while enabling global consistency, we propose an anchor regularizer that aligns user posteriors with item embeddings, stabilizing users under masking and facilitating signal sharing across related items. Our analyses are validated on the Netflix, MovieLens-20M, and Million Song datasets. We also successfully deployed our proposed algorithm on an Amazon streaming platform following a successful online experiment.