Reptile: a Scalable Metalearning Algorithm

By Alex Nichol, John Schulman. 2018

665 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Meta & Continual Learning

Citations in the corpus

Fast and Flexible Multi-Task Classification Using Conditional Neural Adaptive Processes (2019, methodology): “Alternatively, we can adapt all parameters in the feature extractor [7, 8] thereby increasing ∗Authors contributed equally”
Retrieve, Program, Repeat: Complex Knowledge Base Question Answering via Alternate Meta-learning (2020, positive): “In the programmer learning stage, we employed Reptile [Nichol and Schulman, 2018] for fast and simple implementation of MAML while avoiding the significant expense of computing second-order derivatives.”
Meta-Learning with Latent Embedding Optimization (2018, methodology): “While MAML backpropagates the meta-loss through the “inner loop”, REPTILE simpliﬁes the computation by incorporating an L 2 loss which updates the meta-model parameters towards the instance-speciﬁc adapted models.”
Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (2020, methodology): “The model architecture and all other hyperparameters were chosen to match those speciﬁed for Reptile training and evaluation on 1-shot and 5-shot mini-ImageNet in (Nichol & Schulman, 2018).”
A Survey of Deep Network Solutions for Learning Control in Robotics: From Reinforcement to Imitation (2016, neutral): “• Meta-learning : Finn et al. (2017a) and Nichol and Schulman (2018) propose methods that can effectively lead the policy search to ﬁnd parameters that can be adapted to give good performance on a new task with only a small number of training examples of the new task.”
Probabilistic Model-Agnostic Meta-Learning (2018, neutral): “83% Reptile [28] 49 .”
Transfer without Forgetting (2022, background): “MER [47] combines experience-replay and meta-learning [17,40] to maximize transfer from the past while minimizing interference.”
Recent Advances of Continual Learning in Computer Vision: An Overview (2021, positive): “[179] integrated a modified version of the Reptile algorithm [180] into an experience replaymethod to effectively mitigate catastrophic forgetting and preserve the ability of future learning.”
Few-Shot Class-Incremental Learning via Relation Knowledge Distillation (2021, background): “Meta-learning based approaches (Finn, Abbeel, and Levine 2017; Nichol and Schulman 2018): it typically involves an individual meta-learner model that is given a few novel training samples of a new task and tries to quickly learn a learner model that “solves” this new task.”
Towards Continual Reinforcement Learning: A Review and Perspectives (2020, background): “Recently popular approaches for ”meta-learning” (Schmidhuber, 1987; Bengio, Bengio, Cloutier, and Gecsei, 1992) have come to leverage protocols for what is called meta-training and meta-testing that also align with this assumption of a stationary task distribution p(z) (Wang, Kurth-Nelson, Tirumala,”
Continual Learning in Low-rank Orthogonal Subspaces (2020, neutral): “MER [Riemer et al., 2019] is another memory-based method that uses ﬁrst-order meta-learning formulation [Nichol and Schulman, 2018] to reduce forgetting on previous tasks.”
Meta-Consolidation for Continual Learning (2020, methodology): “Meta-learning algorithms - which mostly have focused on the standard learning setting (not continual learning) - can be broadly classified into black-box adaptation based methods [70, 55], optimization based [23, 57, 90] and non-parametric methods [83, 75, 77].”
iTAML: An Incremental Task-Agnostic Meta-learning Approach (2020, methodology): “iTAML is close to Reptile [19] meta-learning algorithm.”