How Does the Brain Form and Find Episodic Memories? A Publish-Subscribe Hypothesis

Memory consolidation remains puzzling partly because cognitive theory often assumes memory to be a passive store under an active processor. This article proposes the publish-subscribe episodic memory hypothesis (PSEMH), modeled on interactive software systems. A hub in association cortex broadcasts a search, and each episodic memory whose guard filter matches it nominates itself for activation. A memory forms in the neocortex at the outset of an episode, while the hippocampus indexes its events. Consolidation moves the memory’s binding from hippocampus to cortex. PSEMH makes testable predictions, not shared by rival accounts, about interference, metabolic cost, re-indexing, and distributed goals. How the brain organizes and finds whole episodes remains unclear, in part because cognitive theory traditionally follows a principle inherited from the Turing machine and the von Neumann architecture: active processors working sequentially on passive memory elements. In computing, this principle creates both a performance and a conceptual bottleneck. Interactive software systems, which map many requests onto many responses without a central controller, avoid it with the publish-subscribe pattern: publishers broadcast without addressing receivers, and subscribers take only what matches their interests. The publish-subscribe episodic memory hypothesis (PSEMH) applies this pattern in five rules. Hubs in association cortex publish search signals (Rule 1). Their subscribers are episodic memories, each with a guard filter for its own combination of components, or cluster hubs for what groups of memories share (Rule 2). A hub passes on only the unmatched remainder of a search, and a memory that accepts it nominates itself for activation (Rule 3). When nothing answers, a new memory is formed in the neocortex at the outset of an episode; the hippocampus indexes each event, and the index is added to the episode’s cortical draft (Rule 4). Each memory owns its goals and anti-goals through outcome marks from a broadcast evaluation signal (Rule 5). Consolidation moves the binding of a memory’s components from hippocampal indexes into binders in the memory’s own cortical body. The hypothesis addresses three problems that current theories leave open and makes ten predictions that separate it from attractor, multiple trace, and central-control accounts.