Mathematical Models of Memory and Forgetting: From Ebbinghaus to Modern Neural Theories

Forgetting Is Not Random

Human experience shows that newly learned items fade quickly while emotionally charged memories persist for decades. In 1885 Hermann Ebbinghaus conducted systematic experiments using meaningless syllables to isolate pure memory decay, measuring retention at multiple intervals over seven months. His data produced the famous Forgetting Curve, showing roughly 42% loss after 20 minutes, one‑third retained after a day, and less than 20% after a month.

Exponential Decay vs. Power‑Law Decay

Ebbinghaus Model

The Ebbinghaus curve can be expressed mathematically as a retention rate R(t) that depends on time t and a stability parameter s; larger s yields a flatter curve, indicating more resistant memory.

When s is small (weakly encoded information), the curve drops steeply; when s is large (repeatedly reviewed material), the decline is gradual.

This is a pure exponential decay model, implying a constant proportional loss of memory over time.

Later research by Wixted & Ebbesen (1991) and Rubin & Wenzel (1996) showed that over longer periods the forgetting curve follows a power‑law function, which decays quickly initially but levels off, matching empirical data better than a simple exponential.

From Behavior to Neural Mechanisms

Memory Chain Model (MCM)

In 2013 Murre, Chessa, and Meeter proposed the Memory Chain Model, assuming that multiple neural systems share two properties: (1) memory strength decays over time, and (2) before complete loss, the memory triggers a new representation in a higher‑level, more stable storage.

In animal studies, “store 1” corresponds to the hippocampus and “store 2” to neocortex. Hippocampal lesions produce a retrograde forgetting gradient that the MCM quantitatively captures, while also fitting normal control data.

ACT‑R Memory Activation Model

John R. Anderson’s ACT‑R framework defines a chunk’s activation as the log‑odds of retrieving that chunk at time t. The basic activation formula includes the time since each practice and a decay parameter d (often set to 0.5).

Pavlik and Anderson later refined the decay mechanism so that the decay rate depends on the current activation: high activation leads to faster decay, low activation yields slower decay, naturally explaining the spacing effect.

Incorporating Individual Differences and Active Forgetting

Traditional models treat forgetting as a passive, unidirectional decay, but newer perspectives view it as an optimization mechanism. Researchers at Southern University of Science and Technology argue that forgetting reorganizes memory, discarding low‑value details to preserve useful patterns, akin to experience replay in reinforcement learning.

A second extension adds physiological variables such as sleep quality, stress, material complexity, and learning strategies, proposing that these factors dynamically modulate the stability parameter rather than remaining constant.

From Models to Applications

The most direct application of memory models is spaced‑repetition software. Piotr Wozniak’s SuperMemo (SM‑2 algorithm, 1987) assigns each item an easiness factor (initially 2.5) and updates it based on a 0‑5 recall rating, computing the next review interval accordingly.

Meta‑analyses report that spaced‑repetition can increase long‑term retention by 200‑300% compared with traditional study methods. The algorithm mathematically combats the Ebbinghaus curve by inserting reviews just before forgetting, gradually increasing the stability parameter.

Key Takeaways

Forgetting has structure. From exponential decay to the Memory Chain Model and ACT‑R activation equations, all models describe predictable, quantifiable patterns.

Each model has limits. The Ebbinghaus curve is simple but coarse; MCM offers neural insight at the cost of many parameters; ACT‑R captures learning dynamics but lacks full treatment of emotion and sleep.

Active forgetting may be crucial. Treating forgetting as a resource‑reallocation process provides valuable insights for education design and AI memory mechanisms.

Understanding these models helps adopt a compassionate view of forgetting as a functional brain process rather than a personal shortcoming.