Spaced repetition memory system

A spaced-repetition memory system combines the Testing effect and the Spacing effect to enable efficient memorization of many thousands of facts (Spaced repetition memory systems are extremely efficient). Some people also use them for a broader set of tasks (see below). Spaced repetition memory systems make memory a choice, but they’re not just for rote facts: Spaced repetition memory systems can be used to develop conceptual understanding.

The first consumer system of this kind was Supermemo, created by Piotr Wozniak. It adopted and popularized the term “spaced repetition”; prior literature had used a variety of terms (often referring to more specific facets of the underlying phenomenon).

• Related systems and possibilities:
  • Differences in my approach to spaced repetition systems
  • Implementations:
  • Traditional: Supermemo, Mnemosyne, Anki
  • Unusual variations:
    • Mnemonic medium
    • Execute Program
    • RemNote
    • Mochi
    • Readwise
    • Chessable MoveTrainer
    • Revuu
  • Unusual applications of spaced repetition memory systems
  • Spaced repetition systems can be used to program attention
  • Spaced repetition systems as catechism
  • Spaced repetition memory systems can be used to prompt application, synthesis, and creation
  • Prefer “memory system” to “spaced repetition system”; Perhaps “practice systems”, not “memory systems”
• On properties of memory systems:
  • What’s the maximum intake rate of an efficient spaced repetition memory system?
  • Optimizing spaced repetition system schedules
  • Self-grading vs. machine-grading in spaced repetition memory systems
  • To what extent does spaced repetition memory system practice constitute deliberate practice?
  • SRS retry mechanism
• Barriers to adoption
  • Many people view memory as unimportant to deep creative work
  • Writing good spaced repetition memory prompts is hard
  • Important attributes of good spaced repetition memory prompts
  • Studying another person’s spaced repetition memory prompts is usually ineffective
  • The limiting factor in spaced repetition system capacity is writing enough good prompts
  • Using machine learning to generate good spaced repetition prompts from explanatory text
  • The critical thing to optimize in spaced repetition memory systems is emotional connection to the review session and its contents
  • Memory system practice sessions are too disconnected from activities you actually care about
  • Regular spaced repetition memory practice is an onerous habit to adopt
    • Spaced repetition memory systems don’t rapidly demonstrate their benefits
    • Spaced repetition memory systems make you feel like your memory is worse than it really is
  • Spaced repetition review sessions often become boring and detached without a steady stream of new prompts
  • The dominant culture around spaced repetition memory systems is fixated on meaningless goals
• Responses to common objections
  • SRS are only relevant for rote learning of facts: “school learning”
  • People underrate the importance of memory because of bad experiences with rote memorization in school
  • Existing SRS are almost exclusively used for simple declarative knowledge, contra Spaced repetition memory systems can be used to develop conceptual understanding
  • Memory systems help you focus on deeper engagement by automating away rote memorization
  • Memory augmentation can accelerate the unpleasant early stages of learning a subject
  • (caveat: it’s hard to write prompts for conceptual understanding!)
  • SRS are only for material you don’t care about enough to remember “naturally.” Better to “learn by doing.”
  • External memory aids are sufficient
  • Mnemonic techniques obviate SRS
  • Knowledge learned through SRS doesn’t transfer
  • The real value in SRS comes from making one’s own prompts
  • Maybe! How important is it to write your own spaced repetition memory prompts?
• On algorithms
  • Simple spaced repetition algorithms mistreat expected failures
  • SRS retry mechanism
  • Optimizing spaced repetition system schedules
• Adoption stories
  • Using spaced repetition with young children

References

Branwen, G. (2009). Spaced Repetition for Efficient Learning. Retrieved December 16, 2019, from https://www.gwern.net/Spaced-repetition

Who invented the name: spaced repetition? - supermemo.guru

What’s the maximum intake rate of an efficient spaced repetition memory system?

Imagine that I add 40 new prompts a day, that I always get all of them right in every review, and that their intervals expand on Quantum Country’s fixed schedule. At the one year mark, I’ll be reviewing:

• prompts from 5 days ago
• prompts added 5+14=19 days ago
• prompts from 19+30=49 days ago
• prompts from 49+60=109 days ago
• prompts from 109+120=229 days ago

So in this instance I’d be reviewing 200 prompts. If a six-second average holds, that’s 20 minutes per day. If we insist on a ten minute limit, that gives me a capacity of 100 prompts, which means about 20 new prompts per day.

Because of the exponent, two years doesn’t change much. If I keep adding 40/day, after adding another ~15k in the following year, I’d be reviewing 240 per day (about 24 minutes). If we want to stick to the ten-minute limit, this would mean 17 new prompts per day. A third year adds another 40 daily prompts under these assumptions; a fourth year adds none.

Of course, this is assuming that I get every answer correct all the time. Assuming a 10% error rate, at the 1-year mark I’d be reviewing 241 prompts instead of 200. Not so bad! This brings my 10-minute capacity from 20/day (7.3k/year to 16/day (5.8k/year). With a 5% error rate, it’s 18/day (around 6.5k/year). The difference is substantial enough to be worth pursuing, but it’s not make-or-break. A 20% error rate yields a 13/day throughput (around 4.8k/year).

So to put it another way, if you can decrease the error rate from 20% to 5%, you can add 1/3 more prompts in the same review period.

Note, though, that Quantum Country’s schedule may not be a good reference for optimality bounds: Quantum Country users rarely forget after demonstrating five-day retention! See also Optimizing spaced repetition system schedules.

const p = 0.9;
const np = 1 - p;
const newCards = 40;
const ease = 2.5;
const max = 365;

const queue = [{baseDays: 5, factor: 1}];
let total = 0;
while (queue.length > 0) {
	const {baseDays, factor} = queue.shift();
	if (baseDays > max || factor < 0.0001) {
		continue;
	}
	total += factor;
	queue.push({baseDays: baseDays * ease, factor: factor * p}, {baseDays: Math.max(5, baseDays / ease), factor: factor * np});
}
console.log(total * newCards);

Q. In a lifetime, roughly how many items can be learned using a 2020-era spaced repetition memory system?
A. $10^6$