FSRS vs SM-2: Free Spaced Repetition Scheduler DSR Model vs SuperMemo Heuristic
Difficulty-Stability-Retrievability memory decay optimization versus legacy interval multiplier
The Free Spaced Repetition Scheduler (FSRS) is a modern machine-learning-optimized spaced repetition algorithm modeling memory decay across three core components: Difficulty (D), Stability (S), and Retrievability (R). SuperMemo-2 (SM-2, developed in 1987) uses an empirical heuristic Ease Factor (EF) multiplier that leads to the notorious 'ease hell' phenomenon and rigid review schedules.
Technical Tradeoff & Benchmark Matrix
| Spaced Repetition Metric | FSRS (DSR Memory Model) | SM-2 (SuperMemo 1987) |
|---|---|---|
| Memory Architecture | Continuous three-variable state: Difficulty, Stability, Retrievability | Single scalar Ease Factor with discrete repetition counts |
| Forgetting Curve Shape | Power-law decay curve backed by empirical cognitive science | Exponential decay assumption with rigid multiplier jumps |
| Ease Hell Vulnerability | Zero (recovering stability adapts without permanently damaging future intervals) | Severe (failing a card repeatedly permanently crushes Ease Factor) |
| Review Workload Efficiency | Reduces total review workload by 15% to 30% for identical retention | Excessive redundant reviews for mature items |
| User Retention Target Tuning | Configurable desired retention rate (e.g. 85%, 90%, 95%) | Fixed arbitrary retention rate (~85%-90%) with no user parameter |
| Parameter Optimization | Custom parameter training over user review history (PyTorch/Rust) | Universal fixed constants hardcoded in 1987 |
Interactive Calculation Engines
Calculate empirical values directly using our zero-latency edge computational models:
Frequently Asked Questions & Technical Analysis
Q: What is 'Ease Hell' in SM-2 and how does FSRS resolve it?
In SM-2, every time a user fails a card, its Ease Factor decreases. After multiple lapses, the Ease Factor drops to minimum (1.3), meaning future review intervals grow at a crawl, forcing hundreds of unnecessary repetitions. FSRS models memory Stability independently of Difficulty, allowing rapid recovery once a concept is relearned.
Q: Why is power-law forgetting more accurate than exponential forgetting in spaced repetition?
Empirical testing across millions of flashcard reviews proves that human memory retrievability declines as a power function of time over prolonged intervals, rather than a steep exponential decay. FSRS models this power law directly.