</> Code with Papers by Syntology

Methods › General › Meta-Learning Algorithms › MAML

Model-Agnostic Meta-Learning

MAML

232 papers tagged archive 2025-07-28

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

MAML, or Model-Agnostic Meta-Learning, is a model and task-agnostic algorithm for meta-learning that trains a model’s parameters such that a small number of gradient updates will lead to fast learning on a new task.

Consider a model represented by a parametrized function f_θ with parameters θ. When adapting to a new task 𝒯ᵢ, the model’s parameters θ become θ′ᵢ. With MAML, the updated parameter vector θ′ᵢ is computed using one or more gradient descent updates on task 𝒯ᵢ. For example, when using one gradient update,

θ′ᵢ = θ- α∇_θℒ_(𝒯ᵢ)(f_θ)

The step size α may be fixed as a hyperparameter or metalearned. The model parameters are trained by optimizing for the performance of f_(θ′ᵢ) with respect to θ across tasks sampled from p(𝒯ᵢ). More concretely the meta-objective is as follows:

min_θ ∑_(𝒯ᵢ ∼p(𝒯)) ℒ_(𝒯ᵢ)(f_(θ′ᵢ)) = ∑_(𝒯ᵢ ∼p(𝒯)) ℒ_(𝒯ᵢ)(f_(θ- α∇_θℒ_(𝒯ᵢ)(f_θ)))

Note that the meta-optimization is performed over the model parameters θ, whereas the objective is computed using the updated model parameters θ′. In effect MAML aims to optimize the model parameters such that one or a small number of gradient steps on a new task will produce maximally effective behavior on that task. The meta-optimization across tasks is performed via stochastic gradient descent (SGD), such that the model parameters θ are updated as follows:

θ←θ- β∇_θ ∑_(𝒯ᵢ ∼p(𝒯)) ℒ_(𝒯ᵢ)(f_(θ′ᵢ))

where β is the meta step size.

Source: Model-Agnostic Meta-Learning for Fast Adaptation of Deep NetworksSee Code · cbfinn/maml

Papers archive 2025-07-28

30 shown of 232, 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 218 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
Meta-Learning208
Few-Shot Learning71
Reinforcement Learning23
Image Classification22
Reinforcement Learning (RL)21
Transfer Learning20
image-classification20
Few-Shot Image Classification17
reinforcement-learning17
General Classification16
Classification12
Diversity11
Federated Learning11
model11
Meta Reinforcement Learning9
Representation Learning9
Domain Adaptation7
regression7
Continual Learning6
Language Modeling6

Usage over time archive 2025-07-28

Papers per year tagged with MAML: 2017 to 2025, peak 61 61 0 2017: 2 papers 2017 2018: 11 papers 2018 2019: 27 papers 2019 2020: 45 papers 2020 2021: 61 papers 2021 2022: 39 papers 2022 2023: 20 papers 2023 2024: 20 papers 2024 2025: 7 papers 2025
Papers per year the archive tags with this method, by the paper's archive date (232 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

Meta-Learning Algorithms

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