Predictive Associative Memory: Retrieval Beyond Similarity Through Temporal Co-occurrence

TL;DR

This paper introduces Predictive Associative Memory (PAM), a neural architecture that learns to recall memories based on temporal co-occurrence rather than similarity, improving associative recall accuracy in synthetic benchmarks.

Contribution

The paper proposes PAM, which uses a JEPA-style predictor trained on temporal co-occurrence, enabling retrieval based on association rather than similarity, a novel approach in memory models.

Findings

97% accurate retrieval of true temporal associates

Achieves cross-boundary Recall@20 of 0.421 with zero cosine similarity

Discriminates experienced-together from never-experienced-together states with AUC of 0.916

Abstract

Current approaches to memory in neural systems rely on similarity-based retrieval: given a query, find the most representationally similar stored state. This assumption -- that useful memories are similar memories -- fails to capture a fundamental property of biological memory: association through temporal co-occurrence. We propose Predictive Associative Memory (PAM), an architecture in which a JEPA-style predictor, trained on temporal co-occurrence within a continuous experience stream, learns to navigate the associative structure of an embedding space. We introduce an Inward JEPA that operates over stored experience (predicting associatively reachable past states) as the complement to the standard Outward JEPA that operates over incoming sensory data (predicting future states). We evaluate PAM as an associative recall system -- testing faithfulness of recall for experienced associations -- rather than as a retrieval system evaluated on generalisation to unseen associations. On a synthetic benchmark, the predictor's top retrieval is a true temporal associate 97% of the time (Association Precision@1 = 0.970); it achieves cross-boundary Recall@20 = 0.421 where cosine similarity scores zero; and it separates experienced-together from never-experienced-together states with a discrimination AUC of 0.916 (cosine: 0.789). Even restricted to cross-room pairs where embedding similarity is uninformative, the predictor achieves AUC = 0.849 (cosine: 0.503, chance). A temporal shuffle control confirms the signal is genuine temporal co-occurrence structure, not embedding geometry: shuffling collapses cross-boundary recall by 90%, replicated across training seeds. All results are stable across seeds (SD < 0.006) and query selections (SD $\leq$ 0.012).