Multivariate Latent Recalibration for Conditional Normalizing Flows

2025-05-22

AI Summary

Modeling the conditional distribution of multivariate response variables remains challenging: existing approaches often suffer from model misspecification or inadequate calibration, leading to inaccurate joint distribution approximations and compromising predictive reliability and decision-making quality. Moreover, mainstream calibration techniques are largely restricted to univariate settings; while predictive intervals may satisfy coverage guarantees, they fail to yield explicit density functions. This paper introduces a novel probabilistic calibration paradigm in latent space: we define a calibration metric within the latent space of conditional normalizing flows and propose the first latent-space recalibration (LR) method, providing an explicit, multivariate density recalibration scheme with finite-sample theoretical guarantees. Our approach integrates conditional normalizing flows, latent-variable modeling, and posterior calibration theory. Experiments on tabular and image data demonstrate that LR significantly reduces latent miscalibration error and negative log-likelihood, achieving a favorable balance among calibration accuracy, explicit density estimation, and computational efficiency.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Reliably characterizing the full conditional distribution of a multivariate response variable given a set of covariates is crucial for trustworthy decision-making. However, misspecified or miscalibrated multivariate models may yield a poor approximation of the joint distribution of the response variables, leading to unreliable predictions and suboptimal decisions. Furthermore, standard recalibration methods are primarily limited to univariate settings, while conformal prediction techniques, despite generating multivariate prediction regions with coverage guarantees, do not provide a full probability density function. We address this gap by first introducing a novel notion of latent calibration, which assesses probabilistic calibration in the latent space of a conditional normalizing flow. Second, we propose latent recalibration (LR), a novel post-hoc model recalibration method that learns a transformation of the latent space with finite-sample bounds on latent calibration. Unlike existing methods, LR produces a recalibrated distribution with an explicit multivariate density function while remaining computationally efficient. Extensive experiments on both tabular and image datasets show that LR consistently improves latent calibration error and the negative log-likelihood of the recalibrated models.

Problem

• Improving multivariate response distribution calibration for reliable predictions
• Extending recalibration methods from univariate to multivariate settings
• Ensuring explicit density function with computational efficiency in recalibration

Innovation

• Introduces latent calibration for normalizing flows
• Proposes LR for post-hoc model recalibration
• Ensures multivariate density with computational efficiency