No Reinstate CMR

No context reinstatement during encoding

No Reinstate CMR (CMR-NoSPR) modifies how context updates when items are re-presented. In standard CMR, each presentation retrieves and reinstates the item’s associated context. This model skips reinstatement during study, using only the pre-experimental item-context mappings.

The Mechanism

Standard CMR updates MFC associations during study. When an item is presented again, these learned associations mean the context input includes:

  1. Pre-experimental context (item’s default unit)
  2. Experimentally-learned context (traces from prior presentations)

No Reinstate CMR separates these: - Context input: Always from pre-experimental MFC (no reinstatement of prior encoding) - Learning: Normal MFC updates (associations still form) - Retrieval: Uses full MFC (with learned associations)

Why No Reinstatement?

The theoretical question: when you see an item again, does it automatically remind you of the first time?

  • Standard CMR: Yes—the item retrieves its encoding context
  • No Reinstate CMR: No—repetitions don’t trigger prior-context retrieval during study

This makes each presentation contextually independent at encoding, while still allowing retrieval-time reinstatement.

Mathematical Specification

Standard CMR (for comparison)

Context input at study uses the full MFC: \[\mathbf{c}^{IN}_i = M^{FC}_{current} \mathbf{f}_i\]

where \(M^{FC}_{current}\) includes learned associations from prior presentations.

No Reinstate CMR

Context input uses only pre-experimental MFC: \[\mathbf{c}^{IN}_i = M^{FC}_{pre} \mathbf{f}_i\]

This returns the item’s default context unit, regardless of prior presentations.

Learning still updates the full MFC: \[\Delta M^{FC}_{ij} = \gamma \mathbf{f}_i \mathbf{c}_j\]

At retrieval, the full MFC is used for context reinstatement.

Implementation

The key difference is maintaining two MFC copies:

Code 
def __init__(self, ...):
    ...
    # Pre-experimental MFC for context updates during study
    self.pre_exp_mfc = mfc_create_fn(list_length, parameters, self.context)
    # Full MFC for learning and retrieval
    self.mfc = mfc_create_fn(list_length, parameters, self.context)

def experience_item(self, item_index):
    item = self.items[item_index]
    # Use pre-experimental MFC for context input
    context_input = self.pre_exp_mfc.probe(item)  # Not self.mfc!
    new_context = self.context.integrate(context_input, self.encoding_drift_rate)

    # But update the full MFC
    return self.replace(
        context=new_context,
        mfc=self.mfc.associate(item, learning_state, self.mfc_learning_rate),
        ...
    )

Predictions

Context Overlap for Repetitions

Standard CMR: Repetition reinstates prior context → overlapping encoding contexts

No Reinstate CMR: Each presentation integrates a fresh context unit → less overlap

This affects: - Spacing: Less natural context overlap between presentations - Contiguity at recall: Recalling an item reinstates context (normal), but encoding didn’t blend contexts

Lag-CRP

At retrieval, context reinstatement works normally. The difference shows in: - Transitions after recalling a repeated item - Whether retrieval cues the first presentation’s neighbors or the second’s

Usage

Code 
from jaxcmr.models.no_reinstate_cmr import CMR

params = {
    "encoding_drift_rate": 0.5,
    "start_drift_rate": 0.5,
    "recall_drift_rate": 0.5,
    "learning_rate": 0.5,
    "primacy_scale": 2.0,
    "primacy_decay": 0.8,
    "shared_support": 0.05,
    "item_support": 0.25,
    "choice_sensitivity": 0.6,
    "stop_probability_scale": 0.05,
    "stop_probability_growth": 0.2,
    "learn_after_context_update": True,
    "allow_repeated_recalls": False,
}

model = CMR(list_length=16, parameters=params)

Comparison: Reinstatement Models

Model Study-time reinstatement Retrieval-time reinstatement
Standard CMR Yes (blends contexts) Yes
No Reinstate CMR No (fresh context each time) Yes
Positional CMR N/A (position-based) Yes (weighted)

Theoretical Implications

This model tests whether:

  1. Automatic retrieval occurs at study
  2. Encoding independence produces different patterns than context blending
  3. Strategic control over reinstatement affects memory

If No Reinstate CMR fits data as well as standard CMR, it suggests that study-time reinstatement may not be necessary to explain the behavioral phenomena.

When To Use

Use No Reinstate CMR when:

  • Testing whether study-time reinstatement matters
  • Modeling conditions that might block automatic retrieval
  • You want encoding contexts to be more independent

Use standard CMR when:

  • Automatic reinstatement is theoretically important
  • You expect context blending to drive behavior