Brain-like Variational Inference

TL;DR

This paper introduces a biologically plausible neural inference framework called FOND, leading to the development of iP-VAE, a recurrent spiking neural network that performs efficient variational inference and outperforms existing models on complex datasets.

Contribution

The paper presents FOND, a principled framework for neural inference based on natural gradients and iterative updates, resulting in the novel iP-VAE model that combines biological plausibility with strong empirical performance.

Findings

iP-VAE outperforms standard VAEs and predictive coding models in sparsity and reconstruction.

iP-VAE scales effectively to complex datasets like CelebA.

iP-VAE generalizes well to out-of-distribution inputs.

Abstract

Inference in both brains and machines can be formalized by optimizing a shared objective: maximizing the evidence lower bound (ELBO) in machine learning, or minimizing variational free energy (F) in neuroscience (ELBO = -F). While this equivalence suggests a unifying framework, it leaves open how inference is implemented in neural systems. Here, we introduce FOND (Free energy Online Natural-gradient Dynamics), a framework that derives neural inference dynamics from three principles: (1) natural gradients on F, (2) online belief updating, and (3) iterative refinement. We apply FOND to derive iP-VAE (iterative Poisson variational autoencoder), a recurrent spiking neural network that performs variational inference through membrane potential dynamics, replacing amortized encoders with iterative inference updates. Theoretically, iP-VAE yields several desirable features such as emergent normalization via lateral competition, and hardware-efficient integer spike count representations. Empirically, iP-VAE outperforms both standard VAEs and Gaussian-based predictive coding models in sparsity, reconstruction, and biological plausibility, and scales to complex color image datasets such as CelebA. iP-VAE also exhibits strong generalization to out-of-distribution inputs, exceeding hybrid iterative-amortized VAEs. These results demonstrate how deriving inference algorithms from first principles can yield concrete architectures that are simultaneously biologically plausible and empirically effective.