CogSci 2025

•

July 31, 2025

•

San Francisco, United States

keywords:

concepts and categories

computational modeling

memory

neural networks

This paper presents a proof of principle for Trace Minimalism (Werning, 2020), a novel philosophical framework for episodic memory. Trace Minimalism claims that remembering does not involve the storage of representational content but rather the reconstruction of past scenarios through the interaction of minimal traces with semantic information. Minimal traces establish a causal link to prior experiences but lack categorical and compositional content. We provide a neuro-computational model using a vector-quantized autoencoder and a transformer-based semantic completion mechanism. Our findings support the hypothesis that remembering is possible without representational memory traces and that minimal traces, in interaction with semantic information, reliably construct past scenarios. The results offer a compelling alternative to classical representational theories of memory while maintaining causal continuity with past experiences.

Downloads

SlidesPaperTranscript English (automatic)

Next from CogSci 2025

Modeling Open-World Cognition as On-Demand Synthesis of Probabilistic Models
technical paper

Modeling Open-World Cognition as On-Demand Synthesis of Probabilistic Models

CogSci 2025

Lionel Wong
Jacob Andreas
+8
Lionel Wong and 10 other authors

31 July 2025

Similar lecture

 Interpretable Traces, Unexpected Outcomes: Investigating the Disconnect in Trace-Based Knowledge Distillation
poster

Interpretable Traces, Unexpected Outcomes: Investigating the Disconnect in Trace-Based Knowledge Distillation

ACL 2026

Siddhant Bhambri
Siddhant Bhambri and 2 other authors

05 July 2026