Skip to content

How Much You Remember Is Decided by How You Test ── One memory, one day later: 71.70 per cent by free recall and 91.07 per cent by savings in relearning ── The gap widens with time, reaching 3.2698 after a month ── And an exponential forgetting curve predicts 10^-13 at a month, breaking down ── [Paper 378]

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)

Abstract

A forgetting curve shows memory falling with time. This paper asks what quantity is falling──the answer is a different quantity for each way of testing, giving three answers one day later. No new mathematical theorem and no new law is claimed. Scope of this paper (scope note): No new mathematical theorem and no new law is claimed──Ebbinghaus’s forgetting curve, the savings method, and power-law retention are all standard. We do not build a theory of memory──whether loss occurs in encoding, storage, or retrieval is not entered. It is named and no more. We perform no experiment──no measurement on participants is made. The computation is on a model stated explicitly. We do not derive the exponents──the retention exponent beta=0.30 and the three measure exponents are assumptions. They are set to match the tendency in the literature. Individual differences are not treated──age, material, and amount of study are not the subject. Interventions are not discussed──study methods such as spaced repetition are not the subject. Relation to earlier papers: Paper 358 showed a depth-of-field number fixed by a convention──there the convention was the circle of confusion; here it is the measure. Paper 352 showed a transition temperature moving with the speed of measurement──there too the measurement made the number. Paper 300 showed whether two things share a root is decidable──by that test, no two of the three measures share a root. Paper 355 showed an empirical law naming its own breakdown──here the exponential names its own at a month. What is added is giving retention on three measures at six times, showing the three answers one day later, showing the gap widening with time, using the exponential’s breakdown at a month as a control, and putting the separator on which measure is used. First, there are at least three measures of retention: free recall, recognition, and savings in relearning (Section 2). Second, this is the core of the paper. One day later the answers split into 71.70, 86.45 and 91.07 per cent (Section 2). Third, the gap widens with time, from 1.0142 after an hour to 3.2698 after a month (Section 3). Fourth, an exponential forgetting curve predicts 9.358x10^-14 after a month (Section 4). Fifth, a power law gives 0.356937, which matches experience; the exponential’s breakdown is evidence that forgetting is not exponential (Section 4). Sixth, the separator is which measure is used, not the amount of memory (Section 5). A forgetting curve shows memory falling with time──but the quantity that is falling differs with the way of measuring. There is more than one way to measure “remembering”──producing it unprompted (free recall), identifying it when shown (recognition), or relearning it faster (savings). All three are called retention, and they differ in severity. So one day later there are three answers──71.70, 86.45 and 91.07 per cent. The same memory, the same moment, the same person; only what is measured differs. The spread is 19.37 points, so “only seven-tenths remains” and “nine-tenths remains” describe the same state. The ordering never changes, and at t=0 all three give 1. And the gap widens with time──after an hour it is only 1.0142, so measured immediately every measure agrees; after a month it is 3.2698. 19.2 per cent by free recall and 62.9 by savings ── “almost all forgotten” and “over sixty per cent remains” describing one memory. Short-interval and long-interval studies cannot be set side by side with different measures. The same thing happens on the model’s side──power law and exponential give 0.987828 and 0.959189 after an hour, indistinguishable over short intervals. But after a month the power law gives 0.356937 and the exponential 9.358x10^-14 ── it predicts that nothing whatever survives from a month ago. That is plainly wrong, and its breakdown is evidence that forgetting is not exponential. The same shape as the empirical law naming its own breaking speed in Paper 355. One thing separates them──which measure is used; not the amount of memory. All three rows are about one memory and one state of the brain: there are three numbers because there are three questions. Fix one measure and results become comparable, but “which is the real retention” has no answer ── all three are retention. To be said honestly──beta=0.30 and the three measure exponents are assumptions made here, set to match the tendency in the literature rather than measured. What is shown is the structure that different measures give different answers, and its size. One last thing──“what percentage remains” acquires an answer only once the measure is stated. A report without its measure could be either 71.70 or 91.07. That one number is written does not mean there is one answer. On the making of this work: The ideas and content of this work stem from the author's own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and checking the algebra. Any remaining errors or misinterpretations are solely the author's. Feedback and corrections are sincerely appreciated. ----- 忘却曲線は「時間とともに記憶が減る」ことを示す。本稿が問うのは減っている量が何かである──答えは測り方ごとに別の量であり、一日後の答が三通りある。新しい定理も法則も主張しない。 本稿の射程(射程注記):新しい定理も法則も主張しない──エビングハウスの忘却曲線・節約法・べき則の保持関数はすべて既知である。記憶の理論を作らない──符号化・貯蔵・検索のどこで失われるかには立ち入らない。名前を挙げるにとどめる。実験をしない──被験者による測定は行わない。明示した模型の上で計算する。指数を導かない──保持のべき則の指数 beta=0.30 も、三尺度の指数も仮定である。文献の傾向に合わせて置いた。個人差を扱わない──年齢・材料・学習量による違いは主題ではない。介入を論じない──間隔反復などの学習法は主題ではない。既刊との関係:論文358 は被写界深度の数が約束で決まることを示した──そこでは約束が許容錯乱円だった。ここでは尺度である。論文352 は転移温度が測る速さで動くことを示した──そこでも測定が数を作った。論文300 は同根か別根かが判定できることを示した──三つの尺度は、その基準でどの二つも別根である。論文355 は経験則が自分の破れを指すことを示した──ここでは指数型が一か月後で自分の破れを示す。加えたのは、三尺度の保持率を六つの時刻で数に出したこと、一日後の答が三通りに分かれることを示したこと、開きが時間とともに広がることを示したこと、指数型が一か月後で破れることを対照として示したこと、分離子を「どの尺度で測るか」に置いたことである。 第一に、保持を測る尺度は少なくとも三つある。自由再生・再認・再学習の節約である(第2節)。 第二に、これが本稿の芯である。同じ一日後で、答が 71.70・86.45・91.07 パーセントに分かれる(第2節)。 第三に、開きは時間とともに広がり、一時間後の 1.0142 倍から一か月後の 3.2698 倍になる(第3節)。 第四に、指数型の忘却曲線は一か月後を 9.358x10^-14 と予言する(第4節)。 第五に、べき則なら 0.356937 で、こちらが実感に合う。指数型の破れが、忘却が指数でないことの証拠になる(第4節)。 第六に、分離子は「どの尺度で測るか」であって、記憶の量ではない(第5節)。 忘却曲線は「時間とともに記憶が減る」ことを示す──だが減っている量は、測り方ごとに別の量である。「覚えている」の測り方は一つではない──何も手がかりなく思い出せるか(自由再生)、見せられて分かるか(再認)、覚え直すのが速いか(再学習の節約)。三つとも「保持」と呼ばれ、厳しさが違う。だから一日後の答が三通りある──71.70・86.45・91.07 パーセント。同じ記憶、同じ時刻、同じ人で、違うのは何を測るかだけである。差は 19.37 ポイントあり、「七割しか覚えていない」と「九割覚えている」が同じ状態を指している。順序はどの時刻でも入れ替わらず、t=0 ではどれも 1 になる。そして開きは時間とともに広がる──一時間後は 1.0142 倍しかなく、直後に測ればどの尺度でもほぼ同じ答が出るが、一か月後には 3.2698 倍になる。自由再生で 19.2 パーセント、節約で 62.9 パーセント──「ほとんど忘れた」と「六割以上残っている」が、同じ記憶についての記述である。短期の実験と長期の実験を、尺度が違うまま並べることはできない。模型の側にも同じことが起きる──べき則と指数型は一時間後には 0.987828 と 0.959189 で近く、短時間ではどちらか区別できない。だが一か月後にはべき則が 0.356937、指数型が 9.358x10^-14 で、指数型は「一か月前に覚えたことは一つも残っていない」と予言する。これは明らかに誤りであり、その破れが、忘却が指数ではないことの証拠になる。論文355 で経験則が自分の破れを指したのと同じ形である。分けているものは一つ──どの尺度で測るかであって、記憶の量ではない。三行とも同じ記憶についてであり、測っている脳の状態は一つ。数が三つあるのは、問いが三つあるからである。尺度を一つに固定すれば比較できるが、「どれが本当の保持か」という問いには答が無い──三つとも保持である。正直に言えば──beta=0.30 も三尺度の指数も本稿が置いた仮定であり、文献の傾向に合わせただけで測定値ではない。示したのは、尺度が違えば答が違うという構造と、その大きさである。最後に一つ──「何パーセント残っているか」という問いは、尺度を書いて初めて答をもつ。尺度を書かない報告は、71.70 と 91.07 のどちらでもありうる。数が一つ書いてあることは、答が一つであることを意味しない。 作成にあたって:本稿の着想と内容は、著者自身の考察に基づくものです。文章の構成整理や英訳、数式の確認には AI(大規模言語モデル)の助力を得ました。最終的な内容の解釈や誤りがあれば、それらはすべて著者の責に帰します。お気づきの点があれば、ご教示いただければ幸いです。

View source

Similar papers

#computer vision Open access Jun 2016

Software Development in Startup Companies: The Greenfield Startup Model

The results are packaged in the Greenfield Startup Model (GSM), which explains the priority of startups to release the product as quickly as possible, and the need to shorten time-to-market, by speeding up the development through low-precision engineering activities.

Carmine Giardino, Nicolò Paternoster, M. Unterkalmsteiner et al. · 178 citations · ⚡14
#computer vision Open access Oct 2016

Software Startups - A Research Agenda

Software startup companies develop innovative, software-intensive products within limited timeframes and with few resources, searching for sustainable and scalable business models.

M. Unterkalmsteiner, P. Abrahamsson, Xiaofeng Wang et al. · 157 citations · ⚡17
#machine learning Review Open access Oct 2016

“Failures” to be celebrated: an analysis of major pivots of software startups

This study conducts a case survey study based on the secondary data of the major pivots happened in 49 software startups, and demonstrates that customer need pivot is the most common among all pivot types.

Sohaib Shahid Bajwa, Xiaofeng Wang, Anh Nguyen-Duc et al. · 127 citations · ⚡15
#computer vision Review Open access May 2015

A survey study on major technical barriers affecting the decision to adopt cloud services

The comparison of adopter and non-adopter sample reveals three potential adoption inhibitor, security, data privacy, and portability, which underlines the importance of the technical and security perspectives for research investigating the adoption of technology.

Nattakarn Phaphoom, Xiaofeng Wang, S. Samuel et al. · 111 citations · ⚡8
#computer vision Conference Open access Dec 2013

Affordable and Energy-Efficient Cloud Computing Clusters: The Bolzano Raspberry Pi Cloud Cluster Experiment

The ongoing work building a Raspberry Pi cluster consisting of 300 nodes is presented, with potential use cases being an inexpensive and green test bed for cloud computing research and a robust and mobile data center for operating in adverse environments.

P. Abrahamsson, S. Helmer, Nattakarn Phaphoom et al. · 110 citations · ⚡7
#computer vision Book Open access Mar 2017

On the Unhappiness of Software Developers

The results indicate that software developers are a slightly happy population, but the need for limiting the unhappiness of developers remains, and 219 factors representing causes of unhappiness while developing software are identified.

D. Graziotin, Fabian Fagerholm, Xiaofeng Wang et al. · 84 citations · ⚡6

Related blog posts

MIT News · Artificial Intelligence Sep 14, 2026

New method enables AI for safety-critical situations

The “HardFlow” algorithm could help generative AI models produce high-quality outputs that obey strict requirements when “pretty close” doesn’t cut it.

GPT-Lab Sep 10, 2026

Responsible AI Must Consider Its Afterlife

AI may appear weightless, but every model depends on physical infrastructure. To understand responsible AI, we need to look beyond algorithms and consider the entire lifecycle of the hardware behind them. The post Responsible AI Must Consider Its Afterlife appeared first on GPT-Lab.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.