aiqbench analysis
This paper introduces Empirical Variational Autoencoder (EVA), a generative framework that improves upon standard VAEs by learning autoregressive latent priors to reduce distribution gaps in sequential data. By adding only a single linear layer, EVA achieves high-fidelity image and sound synthesis with inference speeds competitive against diffusion models. The work advances the efficiency of continuous-valued sequence generation without relying on vector quantization.
Empirical Variational Autoencoder
arXiv:2610.06545v1 Announce Type: new Abstract: We present Empirical Variational Autoencoder, a general generative framework for continuous-valued (i.e., non-vector-quantized) sequences. EVA is based on the evidence lower bound of the Variational Autoencoder (VAE) but learns autoregressive latent priors empirically from training data, which can be implemented only by an additional single linear layer on top of VAEs. By replacing the conventional standard-Gaussian constraint with the self-predicted priors, EVA significantly alleviates the latent distribution gap between prior and posterior which is typically observed in conventional VAEs, and leads to high-fidelity ancestral sampling for sequential data generation. Extensive experiments on image and sound synthesis demonstrate that EVA achieves competitive generation quality with autoregressive diffusion baselines despite its much faster inference time.
Abstract:We present Empirical Variational Autoencoder, a general generative framework for continuous-valued (i.e., non-vector-quantized) sequences. EVA is based on the evidence lower bound of the Variational Autoencoder (VAE) but learns autoregressive latent priors empirically from training data, which can be implemented only by an additional single linear layer on top of VAEs. By replacing the conventional standard-Gaussian constraint with the self-predicted priors, EVA significantly alleviates the latent distribution gap between prior and posterior which is typically observed in conventional VAEs, and leads to high-fidelity ancestral sampling for sequential data generation. Extensive experiments on image and sound synthesis demonstrate that EVA achieves competitive generation quality with autoregressive diffusion baselines despite its much faster inference time.
Machine Learning (cs.LG)
arXiv:2610.06545 [cs.LG]
(or arXiv:2610.06545v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.06545 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 207 words
Related coverage
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
Variational autoencoders with latent high-dimensional steady geometric flows for dynamics
arXiv · Jun 17, 2026
Learning Disentangled Representations with Quantum Variational Autoencoders
arXiv · Oct 7, 2026
Disentanglement with Holographic Reduced Representations
arXiv · Jun 9, 2026