What question did this study set out to answer?

The research aims to develop a memory architecture that minimizes forgetting in AI systems during learning.

March 25, 20260 citationsOpen Access

Memory for All: SAGE — Spatial Associative Geometric Embeddings

Key Points

The research aims to develop a memory architecture that minimizes forgetting in AI systems during learning.
Introduced SAGE, storing knowledge in 3D coordinates rather than weights.
Utilized cosine similarity for retrieval and local Hebbian updates for learning.
Introduced various components including SAGESequenceCube and MultiCube for enhanced memory management.
Demonstrated 92% less forgetting compared to traditional neural networks.
Achieved perfect retention across 200 learning steps.
Reported 58.3% accuracy in word analogy tasks, significantly closing the performance gap with GloVe.

Abstract

Every time an AI system needs to learn something new, it must retrain from scratch — consuming months of compute and significant resources, while still forgetting what it knew before. This paper introduces SAGE — Spatial Associative Geometric Embeddings — a memory architecture that stores knowledge as coordinate positions in a 3D geometric cube rather than in weights. Retrieval uses cosine similarity. Learning uses local Hebbian updates. No backpropagation is required at any stage. Six contributions are made: weight-free 3D coordinate storage; self-organising anti-collision; SAGEDivided working memory with fixed spatial partition; MultiCube horizontal scaling; a hippocampal-inspired consolidation pathway; and SAGESequenceCube — explicit geometric transition memory achieving 100% rollout accuracy — all without backpropagation. Empirical results demonstrate 92% less forgetting than neural networks (0.012 vs 0.150), perfect retention across 200 continuous learning steps, 0.000% sparsity activation, and 58.3% word analogy accuracy with direction training — closing 62% of the gap with GloVe (83.3%).

KI fragen

Bookmark

View Full Paper