Exponential Forgetting Curve with Item Difficulty

Theory · All theories
Evidence · 8 claims (5 for, 1 mixed, 2 against) · 3 studies (2 causal, 1 design), q2–q3 · 1 of 3 report an effect size · 8 claims rest on one study

Description

The article adopts a variant of the exponential forgetting curve, in which recall is binary and recall probability decays exponentially with time since last review at a rate reduced by memory strength. It adds that "there is a constant, item-specific component of the memory decay rate" through a difficulty parameter, and its final form uses Leitner deck position as strength with a global difficulty.

Design Implications

Context

Requirements

• Log data giving the delay since last review and the item's Leitner deck position (or review count) for each interaction.
• Recall outcomes discretized to binary; in the Mnemosyne data, self-reported grades of 2 or higher count as recall.

Constraints

• The authors say the choice between item-specific and global difficulty is less clear from the data; the global form is chosen for practicality and tractability.
• The authors call for more sophisticated memory models that more accurately predict the effect of a review schedule on retention.

Target Learners

• Users of flashcard software such as Mnemosyne

Target Learning Objectives

• Predicting recall of flashcard items
• Retention of reviewed items over time

Claims

• Delay Term Improves Exponential Forgetting Curve Recall Prediction [+M]
• Leitner Deck Position Predicts Recall Better Than Review Count [+M]
• Exponential Forgetting Models With Delay Perform Comparably To 1Pl Irt [+M]
• Item Specific Difficulty Helps Only At Low And High Leitner Decks [~M]
• Memory-decay-based models fit fact-learning datasets better than models insensitive to memory decay [+W]
• Models able to weight performance by recency fit better on the Assistments and KDD datasets, without explicit memory-decay terms being necessary [-W]
• For Chinese tone learning, simpler models insensitive to memory decay fit as well as or better than memory-feature models [-W]
• The retention-interval main effect shows forgetting occurred over one week in both experiments [+W]

The evidence by question

• Does a learner model that tracks memory decay since the last review predict recall better than one that ignores it, and for which kinds of material? Memory-decay-based models fit fact-learning datasets better than models insensitive to memory decay [+M]

Related Theories

• Spaced Repetition

Examples

• Learning decomposition: fit exponential learning curves to compare the relative effectiveness of types of pedagogical support

Key Sources

• Reddy, S., Labutov, I., Banerjee, S., & Joachims, T. (2016). Unbounded Human Learning: Optimal Scheduling for Spaced Repetition. KDD ’16, San Francisco, CA, USA. https://doi.org/10.1145/2939672.2939850

Page metadata

Field	Value
type	theory
title	Exponential Forgetting Curve with Item Difficulty
description	The article adopts a variant of the exponential forgetting curve, in which recall is binary and recall probability decays exponentially with time since last review at a rate reduced by memory strength.
status	review
canonical	True
generated	by claude/unspecified on 2026-09-24
trust	unverified — no one has confirmed this page's content
sources	1 — reddy-2016

Frontmatter is the page's machine-readable record — see the Schema & Guide.