# Lagging Inference Networks and Posterior Collapse in Variational Autoencoders

arXiv: https://arxiv.org/abs/1901.05534 · HTML: https://mlsanity.com/paper/1901.05534/

## Key facts

- Cited by 124 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 16 Jan 2019.

## Built on by (most cited first, 30 of 124)

- [CogView: Mastering Text-to-Image Generation via Transformers](https://mlsanity.com/paper/2105.13290/) (2021)
- [Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models](https://mlsanity.com/paper/2103.04922/) (2021)
- [Diffusion-based Generation, Optimization, and Planning in 3D Scenes](https://mlsanity.com/paper/2301.06015/) (2023)
- [Versatile Diffusion: Text, Images and Variations All in One Diffusion Model](https://mlsanity.com/paper/2211.08332/) (2022)
- [Implicit Chain of Thought Reasoning via Knowledge Distillation](https://mlsanity.com/paper/2311.01460/) (2023)
- [Variational Transformer Networks for Layout Generation](https://mlsanity.com/paper/2104.02416/) (2021)
- [Identifiability of deep generative models without auxiliary information](https://mlsanity.com/paper/2206.10044/) (2022)
- [Latent Diffusion Energy-Based Model for Interpretable Text Modeling](https://mlsanity.com/paper/2206.05895/) (2022)
- [Learning Generative Vision Transformer with Energy-Based Latent Space for Saliency Prediction](https://mlsanity.com/paper/2112.13528/) (2021)
- [Posterior Collapse and Latent Variable Non-identifiability](https://mlsanity.com/paper/2301.00537/) (2023)
- [Consistency Regularization for Variational Auto-Encoders](https://mlsanity.com/paper/2105.14859/) (2021)
- [BoDiffusion: Diffusing Sparse Observations for Full-Body Human Motion Synthesis](https://mlsanity.com/paper/2304.11118/) (2023)
- [Transformer-based Conditional Variational Autoencoder for Controllable Story Generation](https://mlsanity.com/paper/2101.00828/) (2021)
- [Learning Fair Representation via Distributional Contrastive Disentanglement](https://mlsanity.com/paper/2206.08743/) (2022)
- [Impossible Distillation: from Low-Quality Model to High-Quality Dataset & Model for Summarization and Paraphrasing](https://mlsanity.com/paper/2305.16635/) (2023)
- [A Critical Look at the Consistency of Causal Estimation With Deep Latent Variable Models](https://mlsanity.com/paper/2102.06648/) (2021)
- [Latent Feature-Guided Diffusion Models for Shadow Removal](https://mlsanity.com/paper/2312.02156/) (2023)
- [CAST: Counterfactual Labels Improve Instruction Following in Vision-Language-Action Models](https://mlsanity.com/paper/2508.13446/) (2025)
- [Robust Disentangled Variational Speech Representation Learning for Zero-shot Voice Conversion](https://mlsanity.com/paper/2203.16705/) (2022)
- [Disentangled Recurrent Wasserstein Autoencoder](https://mlsanity.com/paper/2101.07496/) (2021)
- [Generative Chemical Transformer: Neural Machine Learning of Molecular Geometric Structures from Chemical Language via Attention](https://mlsanity.com/paper/2103.00213/) (2021)
- [Fuse It More Deeply! A Variational Transformer with Layer-Wise Latent Variable Inference for Text Generation](https://mlsanity.com/paper/2207.06130/) (2022)
- [Language Models As Semantic Indexers](https://mlsanity.com/paper/2310.07815/) (2023)
- [A Personalized Dialogue Generator with Implicit User Persona Detection](https://mlsanity.com/paper/2204.07372/) (2022)
- [Complexity Matters: Rethinking the Latent Space for Generative Modeling](https://mlsanity.com/paper/2307.08283/) (2023)
- [Generative Transformer for Accurate and Reliable Salient Object Detection](https://mlsanity.com/paper/2104.10127/) (2021)
- [Language modeling via stochastic processes](https://mlsanity.com/paper/2203.11370/) (2022)
- [Multi-Rate VAE: Train Once, Get the Full Rate-Distortion Curve](https://mlsanity.com/paper/2212.03905/) (2022)
- [Diverse 3D Hand Gesture Prediction from Body Dynamics by Bilateral Hand Disentanglement](https://mlsanity.com/paper/2303.01765/) (2023)
- [Structured Reordering for Modeling Latent Alignments in Sequence Transduction](https://mlsanity.com/paper/2106.03257/) (2021)

## Abstract (arXiv)

The variational autoencoder (VAE) is a popular combination of deep latent variable model and accompanying variational learning technique. By using a neural inference network to approximate the model's posterior on latent variables, VAEs efficiently parameterize a lower bound on marginal data likelihood that can be optimized directly via gradient methods. In practice, however, VAE training often results in a degenerate local optimum known as "posterior collapse" where the model learns to ignore the latent variable and the approximate posterior mimics the prior. In this paper, we investigate posterior collapse from the perspective of training dynamics. We find that during the initial stages of training the inference network fails to approximate the model's true posterior, which is a moving target. As a result, the model is encouraged to ignore the latent encoding and posterior collapse occurs. Based on this observation, we propose an extremely simple modification to VAE training to reduce inference lag: depending on the model's current mutual information between latent variable and observation, we aggressively optimize the inference network before performing each model update. Despite introducing neither new model components nor significant complexity over basic VAE, our approach is able to avoid the problem of collapse that has plagued a large amount of previous work. Empirically, our approach outperforms strong autoregressive baselines on text and image benchmarks in terms of held-out likelihood, and is competitive with more complex techniques for avoiding collapse while being substantially faster.