Persistent structural memory in neural networks is fundamentally limited by the instability of jointly-learned coordinate systems.
March 25, 2026
Original Paper
The Coordinate System Problem in Persistent Structural Memory for Neural Architectures
arXiv · 2603.22858
The Takeaway
The paper identifies that learned latent embeddings are inherently unstable for long-term memory, leading to 'coordinate incompatibility' during transfer. By switching to fixed random Fourier features as extrinsic coordinates, they demonstrate a way to achieve stable, transferable structural memory that outperforms standard Transformers.
From the abstract
We introduce the Dual-View Pheromone Pathway Network (DPPN), an architecture that routes sparse attention through a persistent pheromone field over latent slot transitions, and use it to discover two independent requirements for persistent structural memory in neural networks. Through five progressively refined experiments using up to 10 seeds per condition across 5 model variants and 4 transfer targets, we identify a core principle: persistent memory requires a stable coordinate system, and any