What is Memory? A Homological Perspective

TL;DR

The paper proposes a delta-homology framework in which memory emerges from the closure of reproducible, delay-locked spike sequences as nontrivial homology generators $[\gamma] \in H_1(\mathcal{Z})$ on a latent cognitive manifold. It develops a spatiotemporal complex $\mathcal{K}_\delta$ and a cell-poset formalism to capture PNGs as persistent 1-cycles, augmented by delta-homology, filtrations, and coherent sheaves to ensure context–content coherence. Central organizational principles—Context-Content Uncertainty Principle (CCUP) and Structure-before-Specificity (SbS)—couple synchronization and recurrence to drive robust memory consolidation and recall, realized through a memory-amortized inference process over topological structures. The model unifies memory with perception and prediction as a dual homology–cohomology loop, offering a non-Turing computation perspective in which memory retrieval reduces to completing prevalidated topological cycles. Biologically, nested hippocampal-cortical loops exemplify the consolidation of transient traces into persistent cycles, enabling efficient recall and compositional memory. These insights suggest new directions for cognitive computing that exploit topological invariants and cycle-closure as foundational computational primitives.

Abstract

We introduce the delta-homology model of memory, a unified framework in which recall, learning, and prediction emerge from cycle closure, the completion of topologically constrained trajectories within the brain's latent manifold. A Dirac-like memory trace corresponds to a nontrivial homology generator, representing a sparse, irreducible attractor that reactivates only when inference trajectories close upon themselves. In this view, memory is not a static attractor landscape but a topological process of recurrence, where structure arises through the stabilization of closed loops. Building on this principle, we represent spike-timing dynamics as spatiotemporal complexes, in which temporally consistent transitions among neurons form chain complexes supporting persistent activation cycles. These cycles are organized into cell posets, compact causal representations that encode overlapping and compositional memory traces. Within this construction, learning and recall correspond to cycle closure under contextual modulation: inference trajectories stabilize into nontrivial homology classes when both local synchrony (context) and global recurrence (content) are satisfied. We formalize this mechanism through the Context-Content Uncertainty Principle (CCUP), which states that cognition minimizes joint uncertainty between a high-entropy context variable and a low-entropy content variable. Synchronization acts as a context filter selecting coherent subnetworks, while recurrence acts as a content filter validating nontrivial cycles.

Figures (9)

• Figure 1: Polychronous memory as topological cycles. (a) A single delay-consistent 1-cycle in the spatiotemporal complex $\mathcal{K}_\delta$. (b) Composition by increasing timing tolerance: two local PNGs merge into a larger nontrivial loop at $\delta' > \delta$, illustrating filtration-driven cycle emergence.
• Figure 2: A portion of the spatiotemporal complex $\mathcal{K}_\Delta^\delta$ with vertices $v_i \in C_0$, temporally consistent edges $e_j \in C_1$, and higher-order timing motif (triangle) $f \in C_2$. The boundary operators $\partial_2$ and $\partial_1$ map 2-chains to their edges and 1-chains to their vertices, respectively. Nontrivial 1-cycles like $e_1 + e_2 + e_3$ generate $H_1(\mathcal{K}_\Delta^\delta)$.
• Figure 3: Cell-poset homology and persistent memory. (a) A 1-cycle is trivialized when a 2-cell $f$ fills the loop, making it a boundary. (b) Without a 2-cell (the boundary of a higher-order cell), the loop remains a nontrivial homology class, representing a persistent, path-dependent memory trace.
• Figure 4: Comparison of (a) Generalized Hough Transform (GHT) for generic shape detection and (b) Delta-Homology Voting in latent cycle-closure inference. In GHT, edge features vote toward a known geometric template (e.g., a car) via a reference point accumulator. In delta–homology, each inference step emits a delta-like activation along a latent cycle $\gamma$, which triggers a persistent memory only when the full topological loop is reconstructed.
• Figure 5: Persistence barcode of a spatiotemporal complex under increasing timing tolerance $\delta$. The blue bar indicates a 1-cycle (corresponding to a polychronous group) that is born at $\delta_1$, persists through $\delta_2$, and dies at $\delta_3$. Bars in gray represent connected components in $H_0$.

Theorems & Definitions (46)

• Definition 1: Spatiotemporal Complex of Neural Activity
• Lemma 1: PNGs as 1-Cycles in the Spatiotemporal Complex
• Lemma 2: PNG Repeatability Induces Homological Cycles
• Lemma 3: Composition of Causal Event Chains
• Definition 2: Filtered Spatiotemporal Complex $\mathcal{K}_\Delta^\delta$
• Example 1: Homology Generator from Recurrent Spiking Pattern
• Definition 3: Cell Poset
• Definition 4: Cell Poset as Causal Scaffold
• Lemma 4: Cycle Closure in the Cell Poset
• Remark 1: Functional Role