Semantic Interaction Information mediates compositional generalization in latent space

• URL: http://arxiv.org/abs/2603.27134v2
• Date: Wed, 01 Apr 2026 22:46:45 GMT
• Title: Semantic Interaction Information mediates compositional generalization in latent space
• Abstract summary: We develop the Cognitive Gridworld, a stationary Partially Observable Markov Decision Process (POMDP)<n> observations are generated jointly by multiple latent variables, yet feedback is provided for only a single goal variable.<n>We analyze Recurrent Neural Networks (RNNs) provided with these interactions, finding that Semantic Interaction Information (SII) explains the accuracy gap between Echo State and Fully Trained networks.
• Score: 0.0
• License: http://creativecommons.org/publicdomain/zero/1.0/
• Abstract: Are there still barriers to generalization once all relevant variables are known? We address this question via a framework that casts compositional generalization as a variational inference problem over latent variables with parametric interactions. To explore this, we develop the Cognitive Gridworld, a stationary Partially Observable Markov Decision Process (POMDP) where observations are generated jointly by multiple latent variables, yet feedback is provided for only a single goal variable. This setting allows us to define Semantic Interaction Information (SII): a metric measuring the contribution of latent variable interactions to task performance. Using SII, we analyze Recurrent Neural Networks (RNNs) provided with these interactions, finding that SII explains the accuracy gap between Echo State and Fully Trained networks. Our analysis also uncovers a theoretically predicted failure mode where confidence decouples from accuracy, suggesting that utilizing interactions between relevant variables is a non-trivial capability. We then address a harder regime where the interactions must be learned by an embedding model. Learning how latent variables interact requires accurate inference, yet accurate inference depends on knowing those interactions. The Cognitive Gridworld reveals this circular dependence as a core challenge for continual meta-learning. We approach this dilemma via Representation Classification Chains (RCCs), a JEPA-style architecture that disentangles these processes: variable inference and variable embeddings are learned by separate modules through Reinforcement Learning and self-supervised learning, respectively. Lastly, we demonstrate that RCCs facilitate compositional generalization to novel combinations of relevant variables. Together, these results establish a grounded setting for evaluating goal-directed generalist agents.

Related papers

• An Introduction to Causal Reinforcement Learning [58.680653905480284]
  Causal inference and reinforcement learning operate over different aspects of the same building block, counterfactual relations.<n>We put under a unifying treatment different modes of learning, including online, off-policy, and causal calculus learning.<n>Specifically, we introduce and discuss through causal lenses generalized policy learning, where to intervene, imitation learning, and counterfactual learning.
  arXiv  Detail & Related papers  (2026-06-23T05:28:33Z)
• Learning Divergence Fields for Shift-Robust Graph Representations [73.11818515795761]
  In this work, we propose a geometric diffusion model with learnable divergence fields for the challenging problem with interdependent data. We derive a new learning objective through causal inference, which can guide the model to learn generalizable patterns of interdependence that are insensitive across domains.
  arXiv  Detail & Related papers  (2024-06-07T14:29:21Z)
• Discovering and Reasoning of Causality in the Hidden World with Large Language Models [109.62442253177376]
  We develop a new framework termed Causal representatiOn AssistanT (COAT) to propose useful measured variables for causal discovery.<n>Instead of directly inferring causality with Large language models (LLMs), COAT constructs feedback from intermediate causal discovery results to LLMs to refine the proposed variables.
  arXiv  Detail & Related papers  (2024-02-06T12:18:54Z)
• Deep Learning-based Group Causal Inference in Multivariate Time-series [8.073449277052495]
  Causal inference in a nonlinear system of multivariate timeseries is instrumental in disentangling the intricate web of relationships among variables. In this work, we test model invariance by group-level interventions on the trained deep networks to infer causal direction in groups of variables.
  arXiv  Detail & Related papers  (2024-01-16T14:19:28Z)
• iSCAN: Identifying Causal Mechanism Shifts among Nonlinear Additive Noise Models [48.33685559041322]
  This paper focuses on identifying the causal mechanism shifts in two or more related datasets over the same set of variables. Code implementing the proposed method is open-source and publicly available at https://github.com/kevinsbello/iSCAN.
  arXiv  Detail & Related papers  (2023-06-30T01:48:11Z)
• On the Joint Interaction of Models, Data, and Features [82.60073661644435]
  We introduce a new tool, the interaction tensor, for empirically analyzing the interaction between data and model through features. Based on these observations, we propose a conceptual framework for feature learning. Under this framework, the expected accuracy for a single hypothesis and agreement for a pair of hypotheses can both be derived in closed-form.
  arXiv  Detail & Related papers  (2023-06-07T21:35:26Z)
• Score-based Causal Representation Learning with Interventions [54.735484409244386]
  This paper studies the causal representation learning problem when latent causal variables are observed indirectly. The objectives are: (i) recovering the unknown linear transformation (up to scaling) and (ii) determining the directed acyclic graph (DAG) underlying the latent variables.
  arXiv  Detail & Related papers  (2023-01-19T18:39:48Z)
• DIDER: Discovering Interpretable Dynamically Evolving Relations [14.69985920418015]
  This paper introduces DIDER, Discovering Interpretable Dynamically Evolving Relations, a generic end-to-end interaction modeling framework with intrinsic interpretability. We evaluate DIDER on both synthetic and real-world datasets.
  arXiv  Detail & Related papers  (2022-08-22T20:55:56Z)
• Disentanglement Analysis with Partial Information Decomposition [31.56299813238937]
  disentangled representations aim at reversing the process by mapping data to multiple random variables that individually capture distinct generative factors. Current disentanglement metrics are designed to measure the concentration, e.g., absolute deviation, variance, or entropy, of each variable conditioned by each generative factor. In this work, we use the Partial Information Decomposition framework to evaluate information sharing between more than two variables, and build a framework, including a new disentanglement metric.
  arXiv  Detail & Related papers  (2021-08-31T11:09:40Z)
• Disentangling Observed Causal Effects from Latent Confounders using Method of Moments [67.27068846108047]
  We provide guarantees on identifiability and learnability under mild assumptions. We develop efficient algorithms based on coupled tensor decomposition with linear constraints to obtain scalable and guaranteed solutions.
  arXiv  Detail & Related papers  (2021-01-17T07:48:45Z)
• Visual Neural Decomposition to Explain Multivariate Data Sets [13.117139248511783]
  Investigating relationships between variables in multi-dimensional data sets is a common task for data analysts and engineers. We propose a novel approach to visualize correlations between input variables and a target output variable that scales to hundreds of variables.
  arXiv  Detail & Related papers  (2020-09-11T15:53:37Z)
• NestedVAE: Isolating Common Factors via Weak Supervision [45.366986365879505]
  We identify the connection between the task of bias reduction and that of isolating factors common between domains. To isolate the common factors we combine the theory of deep latent variable models with information bottleneck theory. Two outer VAEs with shared weights attempt to reconstruct the input and infer a latent space, whilst a nested VAE attempts to reconstruct the latent representation of one image, from the latent representation of its paired image.
  arXiv  Detail & Related papers  (2020-02-26T15:49:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.