Variational Inference
A core method for approximate Bayesian inference that reformulates posterior computation as an optimization problem, approximating the true posterior with a parameterized distribution by maximizing the evidence lower bound (ELBO). Mean-field variational inference, which assumes independence among parameters, is the most commonly used approximation strategy.
Timeline (last 90 days)
On synthetic microgrid trajectories, sampling-based generative heads (lightweight diffusion or flow-matching heads) achieved the best calibration, outperforming variational inference and Monte Carlo Dropout/deep ensembles.
This study proposes a scalable variational inference framework for inference over non-Gaussian spatial fields.
In Bayesian neural networks with continuously increasing width M, the KL divergence regularization term in the ELBO gradually overwhelms the expected log-likelihood term, causing the variational predictive distribution to degenerate into the prior predictive distribution.
Reverse KL divergence is mode-seeking, while forward KL divergence is mean-seeking.
All Facts (4)
On synthetic microgrid trajectories, sampling-based generative heads (lightweight diffusion or flow-matching heads) achieved the best calibration, outperforming variational inference and Monte Carlo Dropout/deep ensembles.
50%UnverifiedThis study proposes a scalable variational inference framework for inference over non-Gaussian spatial fields.
50%UnverifiedIn Bayesian neural networks with continuously increasing width M, the KL divergence regularization term in the ELBO gradually overwhelms the expected log-likelihood term, causing the variational predictive distribution to degenerate into the prior predictive distribution.
50%UnverifiedReverse KL divergence is mode-seeking, while forward KL divergence is mean-seeking.
50%