Model Agnostic Meta Learning Algorithms

[howardyclo]

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (MAML)

• Authors: Chelsea Finn, Pieter Abbeel, Sergey Levine
• Organization: UC Berkeley & OpenAI
• Conference: ICML 2017
• Paper: https://arxiv.org/abs/1703.03400
• Code: https://github.com/cbfinn/maml

On First-Order Meta-Learning Algorithms (Reptile)

• Authors: Alex Nichol and Joshua Achiam and John Schulman
• Organization: OpenAI
• Paper: https://arxiv.org/abs/1803.02999

Probabilistic Model-Agnostic Meta-Learning

• Authors: Chelsea Finn, Kelvin Xu, Sergey Levine
• Organization: UC Berkeley & OpenAI
• Conference: NIPS 2018
• Paper: https://arxiv.org/abs/1806.02817

Bayesian Model-Agnostic Meta-Learning

• Authors: Taesup Kim, Jaesik Yoon, Ousmane Dia, Sungwoong Kim, Yoshua Bengio, Sungjin Ahn
• Organization: Element AI, MILA, SAP, Kakao Brain, CIFAR Senior Fellow, Rutgers University
• Conference: NIPS 2018
• Paper: https://arxiv.org/abs/1806.03836

Meta-Learning with Latent Embedding Optimization (LEO)

• Authors: Andrei A. Rusu, Dushyant Rao, Jakub Sygnowski, Oriol Vinyals, Razvan Pascanu, Simon Osindero & Raia Hadsell
• Organization: DeepMind
• Conference: ICLR 2019
• Paper: https://arxiv.org/abs/1807.05960

How to Train Your MAML (MAML++)

• Authors: Antreas Antoniou, Harrison Edwards, Amos Strokey
• Organization: University of Edinburgh & OpenAI
• Conference: ICLR 2019
• Paper: https://arxiv.org/abs/1810.09502
• Code: https://github.com/AntreasAntoniou/HowToTrainYourMAMLPytorch

[howardyclo]

MAML

• A model-agnostic (only gradient decent) meta learning algorithm aims to find a good initialization point for model such that it can be fine-tuned quickly on new tasks.
• Shows STOA on few-shot image classification, regression and fast fine-tuning for policy gradient.

How does it work?

• Sample a batch of tasks (with few training data and "virtual" test data; the "virtual" test data is constructed from training data).
• For each task_{i}
  • Compute gradient w.r.t L(θ) on training data and update model θ -> θ'_{i}.
  • Compute L(θ'_{i}) on virtual test data.
• Compute gradient w.r.t. Σ_{i} L(θ'_{i}) and update model θ.
• The loss can be cross-entropy loss for classification, MSE for regression and reward for RL.

FOMAML

• They also tried ignoring second derivatives θ'{i} in task{i}, just update θ using Σ_{i} L(θ_{i}) (no need train/test splits in original task setup).
• This is denoted as first-Order MAML (FOMAML).
• Shows comparable to second-order MAML and save more computation time.

[howardyclo]

Reptile

• Introduce a new FOMAML algorithm, Reptile, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task.
• Really worth-reading, especially its analysis on SGD and MAML!

How does it work?

• The U^{k}_{T} means we take k gradient updates in the sampled task T; ϵ = 1/α, i.e. learning rate.
• We can update in batch version (n = number of tasks):
  
• If we only take k = 1 update, this is essentially SGD on the expected loss.
• If we take k > 1 updates, it is not. This will converge differently which considers second-and-higher derivatives.
• Experiment on 5-shot 5-way Omniglot shows different inner-loop gradient combinations:
  

Why does it work?

• Through Taylor expansion analysis, both MAML and Reptile contain the same leading order terms:
  • First: minimizing the expected loss (joint training on different tasks).
  • Second: maximizing within-task generalization, i.e., maximizing the inner product between k gradients from the same task. If gradients from different batch has positive inner product, then taking a gradient step on one batch improves performance on the other batch.
• The result of Taylor expansion on SGD and MAML (i in [1, k]):
  
  
• This explains k = 2 in the above experiment is still insufficient since it puts less weight on the second inner product term relative to the first term.