Instance-Adaptive Parametrization for Amortized Variational Inference
Andrea Pollastro, Andrea Apicella, Francesco Isgrò †, Roberto Prevete † · 2026
Read Instance-Adaptive Parametrization for Amortized Variational Inference by Andrea Pollastro, Andrea Apicella, Francesco Isgrò †, Roberto Prevete † (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-06796.txt.
Description
Pollastro and three colleagues narrow the amortization gap in variational autoencoders. Shared encoder settings make posterior fitting cheap but leave a gap versus per-instance inference. Their IA-VAE adds a hypernetwork that tunes a shared encoder per input, adapting it within a single forward pass. Theory shows the IA-VAE family contains standard amortized methods, so its best bound cannot worsen. Tests on synthetic data with known posteriors and on image benchmarks show more accurate fits and better held-out bounds with far fewer weights.
Catalog record
| CREATOR | Andrea Pollastro, Andrea Apicella, Francesco Isgrò †, Roberto Prevete † |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63408 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.06796