Differentiable Clone-Structured Causal Graphs for End-to-End Cognitive Map Learning from Image Sequences

2026-07-14

AI Summary

This study addresses the challenge of constructing structured cognitive maps solely from continuous raw image sequences and self-motion cues, particularly in perceptually ambiguous environments lacking repetitive patterns. To this end, we propose the first end-to-end differentiable Clone-Structured Causal Graph (gradCSCG), jointly trained with a VQ-VAE frontend to directly learn interpretable cognitive graph structures from raw visual inputs. The method incorporates soft emission probabilities and a loss-balancing mechanism to effectively prevent module collapse. Experiments demonstrate that our approach accurately recovers the underlying adjacency graphs across four highly confounding settings—including symbolic grid worlds and MNIST image sequences—thereby validating its efficacy in building structured cognitive maps from realistic visual observations.

Abstract

How can an agent build a structured map of its world from nothing but an ongoing sequence of raw sensory input and its own movements, especially when natural variation means exact sensory patterns rarely repeat? The Clone-Structured Causal Graph algorithm (CSCG), a normative hippocampus model, shows how an interpretable map can be learned from aliased observations. However, CSCG requires a predefined discrete alphabet, and its expectation-maximization formulation is not easily combined with existing neural network modules, preventing the end-to-end processing of raw image sequences. We remove this barrier by reformulating CSCG as a single, fully differentiable module, gradCSCG, and coupling it to a learned vector-quantized variational autoencoder (VQ-VAE) perceptual front-end. A soft emission forward pass allows the map-learning objective to flow back into perception, while a set of loss-balancing mechanisms mitigates module collapse during joint training. We demonstrate, first, that gradient training reproduces CSCG's results on original symbolic grid worlds by recovering room topology from heavily aliased observations. Second, we show that map recovery remains robust on MNIST image sequences, where each visit to a location yields a newly sampled image of its assigned digit. Across four heavily aliased environments, the end-to-end pipeline successfully uncovers the underlying adjacency graph with high edge precision and recall, directly from visual input. This work provides a proof of principle that CSCG can serve as a composable building block in a deep learning architecture.

Problem

• cognitive map learning
• aliased observations
• structured representation
• end-to-end learning
• causal graph

Innovation

• differentiable causal graph
• end-to-end cognitive mapping
• clone-structured representation
• VQ-VAE
• aliased observation learning