This repository constains Julia code for the Generalized Quantum Episodic Memory (GQEM) model of item recognition. In the recognition memory task, subjects study a list of words. In the test phase, three types of words are presented: old words from the study list, new but semantically related words, and new but unrelated words. Subjects are given four sets of instructions

1. gist: respond yes to semantically related words (G)
2. verbatim: respond yes to old (i.e. studied) words (V)
3. gist + verbatim: respond yes to semantically related and old words (G ∪ V)
4. unrelated: respond yes to unrelated words (U)

The law of total probability is violated in experiments, such that Pr(G) + Pr(V) > Pr(G ∪ V). Similarly, the judgments are subadditive: Pr(G) + Pr(V) + Pr(U) > 1. These effects emerge in the GQEM because the memory representations are incompatible, meaning they are represented with different, non-orthogonal bases and evaluated sequentially. As a result, LOTP and additivity do not necessarily hold.

Installation

To install from the REPL, use ] to switch to the package mode and enter the following:

add https://github.com/itsdfish/QuantumEpisodicMemory.jl

Example

The following example shows a violation of the total law of probability.

using QuantumEpisodicMemory

# basis rotation parameters relative to the standard verbatim basis, V
θG = -.12
θU = -1.54
θψO = -.71
θψR = -.86
θψU = 1.26

dist = GQEM(; θG, θU, θψO, θψR, θψU)
preds = compute_preds(dist)

table = to_table(preds)

# violation of LOPT
sum(table[["gist","verbatim"],:], dims=1) - table["gist+verbatim", :]'

1×3 Named Matrix{Float64}
condition ╲ word type │       old    related  unrelated
──────────────────────┼────────────────────────────────
sum(condition)        │  0.570677   0.419159  0.0797084

References

Trueblood, J. S., & Hemmer, P. (2017). The generalized quantum episodic memory model. Cognitive Science, 41(8), 2089-2125.