Quantum-like memory benchmark reveals narrow advantage over classical controls

A new study by Yashine H. Goolam Hossen, Lea Gassab, and Travis J. A. Craddock presents a benchmark for evaluating context-sensitive associative memory with adaptive plasticity. The work addresses a key challenge in learning and memory models: balancing plasticity (encoding new information) with stability (preserving old structure without saturation). Many associative-memory models appear to succeed because fixed background connectivity already carries part of the task, making it hard to distinguish genuine recall from structural assistance. To isolate true learning dynamics, the benchmark uses an order-sensitive staged recall task under weak-support conditions, comparing a quantum-like model (where 'quantum-like' refers to the modeling formalism, not biological quantum computation) against matched real-valued no-phase and Markov-rate controls.

The results reveal that weak structural support alone does not rescue recall in the absence of plasticity. Most gains come from adaptive plasticity, particularly homeostatic stabilization. Interestingly, the Markov-rate control often achieves stronger raw recall scores, but the quantum-like model consistently preserves order sensitivity and stage-dependent organization better. This demonstrates that model classes are best distinguished by a multi-objective profile — combining recall, temporal organization, and context sensitivity — rather than by any single metric. The benchmark provides a controlled framework for studying context-sensitive memory dynamics under regulated plasticity and matched classical comparisons, showing no universal quantum advantage but important qualitative differences.