Under the mean-recall approximation, the maximum achievable learning rate is convex in the learner's review frequency budget, suggesting increasing returns at lower budgets.¶

Claim · All claims
Evidence · 1 study · theoretical r3 · q1
Currency · newest study 2016 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q1 i? The optimized learning rate plotted against review budget U (n = 5, θ = 0.01) is convex for low U, suggesting increasing returns as budget grows. → Reddy 2016

Evidence¶

Reddy 2016¶

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

q1 · i? · theoretical · r3

Optimization result under the mean-recall approximation (Fig. 11). The convexity "suggests that there are increasing returns (for lower U) as the user increases their budget".

"The convexity of the latter plot is encouraging, as it suggests that there are increasing returns (for lower U) as the user increases their budget."

Discussion¶

Related Claims¶

• In simulation, the mean-recall approximation matches the clocked-delay Leitner Queue Network for small arrival rates, and its phase-transition threshold appears to be a conservative lower bound. — related
• Under the mean-recall approximation, the optimal Leitner Queue Network review schedule spends more time on lower decks than on higher decks. — related
• Under the mean-recall approximation, the optimal Leitner Queue Network schedule increases the expected delay between reviews as an item moves up through the decks. — related
• Under the Leitner Queue Network model, raising the rate of new-item introduction beyond a threshold causes a phase transition in learning rate, which a Mechanical Turk vocabulary experiment reproduced. — related
• Under the Leitner Queue Network optimization, easy items call for roughly uniform time across decks, while more difficult items call for more time on lower decks. — related