</> Code with Papers by Syntology

Methods › Reinforcement Learning › Replay Memory › Prioritized Experience Replay

Prioritized Experience Replay

138 papers tagged archive 2025-07-28

Introduced by Tom Schaul et al. in Prioritized Experience Replay

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

Prioritized Experience Replay is a type of experience replay in reinforcement learning where we more frequently replay transitions with high expected learning progress, as measured by the magnitude of their temporal-difference (TD) error. This prioritization can lead to a loss of diversity, which is alleviated with stochastic prioritization, and introduce bias, which can be corrected with importance sampling.

The stochastic sampling method interpolates between pure greedy prioritization and uniform random sampling. The probability of being sampled is ensured to be monotonic in a transition's priority, while guaranteeing a non-zero probability even for the lowest-priority transition. Concretely, define the probability of sampling transition i as

P(i) = (pᵢ^α)/(∑ₖ pₖ^α)

where pᵢ > 0 is the priority of transition i. The exponent α determines how much prioritization is used, with α=0 corresponding to the uniform case.

Prioritized replay introduces bias because it changes this distribution in an uncontrolled fashion, and therefore changes the solution that the estimates will converge to. We can correct this bias by using importance-sampling (IS) weights:

wᵢ = (1/N·1/(P(i)))^β

that fully compensates for the non-uniform probabilities P(i) if β= 1. These weights can be folded into the Q-learning update by using wᵢδᵢ instead of δᵢ - weighted IS rather than ordinary IS. For stability reasons, we always normalize weights by 1/maxᵢwᵢ so that they only scale the update downwards.

The two types of prioritization are proportional based, where pᵢ = |δᵢ| + ϵ and rank-based, where pᵢ = 1/(rank(i)), the latter where rank(i) is the rank of transition i when the replay memory is sorted according to |δᵢ|, For proportional based, hyperparameters used were α= 0.7, β₀ = 0.5. For the rank-based variant, hyperparameters used were α= 0.6, β₀ = 0.4.

PaperSource

Papers archive 2025-07-28

30 shown of 138, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

20 shown of 82 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Reinforcement Learning (RL)72
reinforcement-learning68
Reinforcement Learning60
Deep Reinforcement Learning41
Model-based Reinforcement Learning20
Atari Games17
Q-Learning17
Continuous Control15
continuous-control14
Decision Making11
Board Games8
OpenAI Gym8
Multi-agent Reinforcement Learning7
Game of Go6
MuJoCo5
Offline RL5
Starcraft5
model5
Autonomous Driving4
Atari Games 100k3

Usage over time archive 2025-07-28

Papers per year tagged with Prioritized Experience Replay: 2015 to 2025, peak 26 26 0 2015: 1 paper 2015 2016: 0 papers 2016 2017: 2 papers 2017 2018: 8 papers 2018 2019: 10 papers 2019 2020: 22 papers 2020 2021: 26 papers 2021 2022: 21 papers 2022 2023: 22 papers 2023 2024: 17 papers 2024 2025: 9 papers 2025
Papers per year the archive tags with this method, by the paper's archive date (138 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Replay Memory

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections