We propose to adopt a causal framework for preference learning to define and address challenges like causal misidentification, preference heterogeneity, and crucially, confounding due to user-specific objectives.

We train LLM agents as Language-conditioned policies without requiring expensive labeled data or online experimentation. The framework leverages LLMs to enable the use of unlabeled datasets and improve generalization to unseen goals and states.

This paper formalizes task ambiguity in tasks specified in natural language and frames task disambiguation through Bayesian Experimental Design, leading to more effective strategies for LLMs to pose clarifying questions.

We introduce the paradigm of informed meta-learning, a novel approach to inductive bias specification based on human knowledge represented in any form, including unstructured natural language.

We introduce new tabular few-shot learning solution capable of knowledge sharing between datasets with heterogenous sets of columns.

Chain event graphs (CEGs) are a recent family of probabilistic graphical models that generalise the popular Bayesian networks (BNs) family. In this paper, we present cegpy—the first Python implementation of CEGs and the first across all languages to support structurally asymmetric processes.