Bayesian Meta-Learning with Expert Feedback for Task-Shift Adaptation through Causal Embeddings

TL;DR

This paper introduces a causally-aware Bayesian meta-learning approach that leverages expert feedback and causal embeddings to improve out-of-distribution task adaptation, reducing negative transfer especially in real-world clinical settings.

Contribution

It presents a novel method that conditions task priors on causal embeddings and incorporates expert judgments, addressing negative transfer in out-of-distribution scenarios.

Findings

Reduces negative transfer in out-of-distribution tasks

Improves adaptation in clinical prediction settings

Theoretically controls prior mismatch using causal embeddings

Abstract

Meta-learning methods perform well on new within-distribution tasks but often fail when adapting to out-of-distribution target tasks, where transfer from source tasks can induce negative transfer. We propose a causally-aware Bayesian meta-learning method, by conditioning task-specific priors on precomputed latent causal task embeddings, enabling transfer based on mechanistic similarity rather than spurious correlations. Our approach explicitly considers realistic deployment settings where access to target-task data is limited, and adaptation relies on noisy (expert-provided) pairwise judgments of causal similarity between source and target tasks. We provide a theoretical analysis showing that conditioning on causal embeddings controls prior mismatch and mitigates negative transfer under task shift. Empirically, we demonstrate reductions in negative transfer and improved out-of-distribution adaptation in both controlled simulations and a large-scale real-world clinical prediction setting for cross-disease transfer, where causal embeddings align with underlying clinical mechanisms.