---
title: Refined α-Divergence Variational Inference (RDVI)
url: https://www.emergentmind.com/topics/refined-divergence-variational-inference-rdvi
type: topic
---

# Refined α-Divergence Variational Inference (RDVI)

Refined α-Divergence Variational Inference (RDVI) is an advanced framework for approximate Bayesian inference that generalizes classic variational inference by replacing the standard Kullback–Leibler objective with the Rényi α-divergence. RDVI further integrates rejection sampling to minimize worst-case density ratio discrepancies between the target and variational distributions, yielding improved empirical and theoretical performance, especially in complex or multimodal posterior landscapes [1909.07627]. This method unifies and extends prior α-divergence-based VI schemes, permitting rigorous control over mode-seeking, mass-covering, and robustness properties through the parameter α, and introduces a two-stage optimization-refinement procedure that is theoretically guaranteed to improve the fit.

## 1. Formal Definition of RDVI and α-Divergence

RDVI is grounded in the minimization of the Rényi α-divergence, which for two densities $p(x)$ (target, possibly unnormalized) and $q_\theta(x)$ (variational approximation), is given as
\[
D_\alpha(p\|q_\theta) = \frac{1}{\alpha-1} \log \int q_\theta(x) \left[ \frac{\tilde{p}(x)}{q_\theta(x)} \right]^\alpha dx - \frac{\alpha}{\alpha-1} \log Z_p,
\]
where $p(x)=\tilde{p}(x)/Z_p$ and $\alpha>0$, $\alpha\neq1$ [1909.07627]. In the limit α→1, this recovers the standard KL divergence. The α-divergence allows continuous control between mass-covering (α<1) and mode-seeking (α>1), and provides a unified variational inference framework [1602.02311].

## 2. Two-Stage RDVI: Hybridization with Rejection Sampling

A key innovation of RDVI is embedding a rejection sampling (RS) refinement stage. The crucial observation is that
\[
\lim_{\alpha\to\infty} D_\alpha(p\|q_\theta) = \log \max_x \frac{p(x)}{q_\theta(x)},
\]
which is equivalent to the log of the optimal rejection-sampler constant $M(\theta)$ for $q_\theta(x)$. This leads to the two-stage α-Divergence Rejection Sampling (α-DRS) algorithm [1909.07627]:
- **Stage 1:** Learn $q_\theta$ by minimizing a Monte Carlo estimate of $D_\alpha(p\|q_\theta)$ for finite $\alpha$, resulting in a proposal distribution well-matched to $p$.
- **Stage 2:** Estimate an (approximate) acceptance threshold $T$ (linked to $D_\infty$), and perform rejection sampling with acceptance probability
\[
a_{\hat\theta}(x\mid T) = \frac{1}{1 + [q_{\hat\theta}(x) e^{-T}/\tilde{p}(x)]},
\]
yields a sample-based refined approximation $r_{\hat\theta}(x)$ that provably reduces the α-divergence to the target.

### Stages in Algorithmic Form

| Stage      | Steps                                                                                                                                          | Output                                   |
|------------|-----------------------------------------------------------------------------------------------------------------------------------------------|------------------------------------------|
| 1 (RDVI)   | Optimize Monte Carlo estimate of $D_\alpha(p\|q_\theta)$ over $\theta$ using gradient-based updates                                           | Proposal $q_{\hat\theta}(x)$             |
| 2 (RS)     | Compute threshold $T$ (quantile or $\log M(\theta)$); draw $x\sim q_{\hat\theta}$ and accept with $a_{\hat\theta}(x|T)$ until sufficient samples collected | Empirical distribution $r_{\hat\theta}$ |

Rigorous analysis guarantees that $D_\alpha(p\|r)\le D_\alpha(p\|q_\theta)$ for any finite α, and in the limit $T\to-\infty$, $r$ converges to $p$ [1909.07627].

## 3. Theoretical Properties and Bounds

RDVI establishes that the RS-based refinement always (strictly) decreases the α-divergence between the target $p$ and the approximate $q_\theta$. As $T$ decreases, the refined distribution $r_{\hat\theta}$ becomes increasingly concentrated on high-density regions of $p$, bridging the gap between variational and sampling-based approaches. The result holds for all α>0 and all proposal choices. In the high-dimensional regime, quantile-based thresholds for $T$ mitigate the curse of dimensionality in naive RS [1909.07627].

## 4. Empirical Performance and Case Studies

Empirical studies in [1909.07627] demonstrate the practical efficacy of the RDVI+RS approach:
- **Synthetic Gaussian Mixture:** After Stage 1, $q_\theta$ may miss certain posterior modes; after RS refinement, the sample set $r_{\hat\theta}$ successfully recovers all target modes, indicated by a stark drop in $D_\alpha(p\|r)$ relative to $D_\alpha(p\|q)$.
- **Bayesian Neural Network Regression (UCI):** On benchmarks, α-DRS achieves lower RMSE and higher log-likelihood than standalone RDVI or other $f$-divergence minimization baselines, especially with γ≈0.1 acceptance rates in RS.

These results highlight that the hybrid scheme yields both tighter fits and improved predictive accuracy compared to variational-only routines.

## 5. Algorithmic Implementation and Optimization Strategy

The practical RDVI+RS routine proceeds as follows [1909.07627]:
1. **Input:** Unnormalized target $\tilde{p}(x)$, α>0, acceptance hyperparameter γ.
2. **Variational phase:** Randomly initialize $\theta$; iteratively sample minibatches from $q_\theta$, estimate the α-divergence, and perform stochastic gradient updates to minimize the divergence.
3. **Threshold estimation:** For low dimension, set $T=-\hat D_\alpha(p\|q_{\hat\theta})$; for high dimension, compute $T$ as the empirical quantile of $\mathcal{L}(x) = -\log\tilde{p}(x) + \log q_{\hat\theta}(x)$.
4. **Refinement:** Repeatedly draw $x\sim q_{\hat\theta}$, accept using $a_{\hat\theta}(x|T)$, and accumulate accepted samples until the empirical sample set matches $p$.

This algorithm is amenable to stochastic optimization, mini-batching, and reparameterization-trick based variance reduction.

## 6. Significance, Comparisons, and Broader Implications

RDVI with rejection sampling serves as a principled, provably improved bridge between purely variational and sampling-based inference. It incorporates the flexibility of α-divergence minimization, enabling continuous interpolation between ELBO/VB, mode-seeking, and mass-covering behaviors [1602.02311, 1511.03243]. The explicit connection to $D_\infty$ endows the method with strong guarantees on worst-case density ratio control, rendering it particularly robust to mode misspecification and tail coverage issues that plague naive VI.

The framework's modularity allows seamless adaptation to high-dimensional and complex latent variable models; it is also extensible to mixture optimization, mirror descent, and neural variational representations, interfacing cleanly with geometric and black-box VI improvements [1707.09714, 2007.03814, 2106.05114, 2103.05684]. 

A plausible implication is that future work can explore further meta-learned or adaptive thresholding in Stage 2, richer proposal families, and joint refinement of both $q_\theta$ and $T$, offering avenues for closing the gap with gold-standard MCMC while retaining variational efficiency.

## 7. Summary Statement

Refined α-Divergence Variational Inference via rejection sampling (α-DRS) leverages the $\alpha \rightarrow \infty$ asymptotics of Rényi divergence and a two-stage procedure to obtain a sample-based approximation that never worsens—and typically sharply improves—the closeness of the variational posterior to the target, compared to standalone VI. This unites advances in mode/covering control, robust optimization, and sample-based refinement into a cohesive, theoretically justified, and empirically validated variational inference framework [1909.07627].

Source: https://www.emergentmind.com/topics/refined-divergence-variational-inference-rdvi