Probabilistic Circuits That Know What They Don't Know

TL;DR

This paper reveals that probabilistic circuits are not inherently robust to out-of-distribution data and introduces tractable dropout inference (TDI), a novel method for efficient uncertainty estimation that enhances their robustness.

Contribution

The paper proposes TDI, a tractable, sampling-free uncertainty estimation method for probabilistic circuits, improving their robustness to distribution shifts and out-of-distribution data.

Findings

TDI provides accurate uncertainty estimates in a single forward pass.

Probabilistic circuits' robustness to OOD data is improved with TDI.

Experiments show enhanced confidence calibration and robustness.

Abstract

Probabilistic circuits (PCs) are models that allow exact and tractable probabilistic inference. In contrast to neural networks, they are often assumed to be well-calibrated and robust to out-of-distribution (OOD) data. In this paper, we show that PCs are in fact not robust to OOD data, i.e., they don't know what they don't know. We then show how this challenge can be overcome by model uncertainty quantification. To this end, we propose tractable dropout inference (TDI), an inference procedure to estimate uncertainty by deriving an analytical solution to Monte Carlo dropout (MCD) through variance propagation. Unlike MCD in neural networks, which comes at the cost of multiple network evaluations, TDI provides tractable sampling-free uncertainty estimates in a single forward pass. TDI improves the robustness of PCs to distribution shift and OOD data, demonstrated through a series of experiments evaluating the classification confidence and uncertainty estimates on real-world data.