Remembering without (representational) memory: A neuro-computational study on regaining categoricity and compositionality from minimal traces

This paper provides a proof of principle for the philosophical theory of Trace Minimalism (Werning, 2020), a novel account of episodic memory. It claims that remembering goes without the storage of representational content in memory. Remembering rather consists in the construction of a representation about a scenario, previously experienced, through the interaction of minimal traces with acquired statistical regularities. A minimal trace merely constitutes a causal link to the experience but possesses no trans-temporally transmitted representational (viz. categorical and compositional) content. Our proof of principle uses a neuro-computational model that is based on an autoencoder modeling the visual pathway and a transformer modeling the semantic information network. It assumes the hippocampus as the host of minimal traces. Our model demonstrates that minimal traces, which lack sufficient information for the reliable categorization and composition of content on their own, can be enhanced through interaction with semantic information. This process allows for the accurate construction of past scenarios, respecting regularities in the world, and supporting the reliability of episodic memory. Our findings suggest that episodic memory relies on sparse, fragmentary information rather than stored content, providing a robust proof of principle that remembering is possible without representational memory traces.