1
Experiment - Regression Introduction Risto Vuorio* Multimodal Model-Agnostic Meta-Learning via Task-Aware Modulation Joseph J. Lim Shao-Hua Sun* Hexiang Hu Our Approach Experiment - Classification Experiment - Reinforcement Learning Experiment - Learned Task Embeddings Point Mass Reacher Ant Modulation Network Task Network x y ( ( K ⇥ Samples Task Encoder υ Task Embedding Modulation Network Modulation Network MLPs x y ✓ 2 ⌧ 2 ✓ 1 ⌧ 1 ⌧ n ✓ n … ˆ y Outer loop • Task Encoder: produce the task embedding • MLPs: modulate the task network blocks Inner loop • Task network: fast adapt through gradient updates Parameters ! h ! g ✓ Intuition • Modulation network: identify task modes and modulate the initialization accordingly • Task network: further gradient adaptation via MAML steps Background Model-Agnostic Meta-Learning [1] • Meta-learn a parameter initialization that can be fine-tuned for new tasks in few gradient update steps • Inner loop Model-Agnostic Meta-Learning Objective • Outer loop [1] Finn, Chelsea, Pieter Abbeel, and Sergey Levine. "Model-agnostic meta-learning for fast adaptation of deep networks." in International Conference on Machine Learning 2017 Unimodal Task Distribution Multimodal Task Distribution Real-world task distributions are often multimodal • Have a rich structure (e.g. multiple modes) • Some knowledge can be transferable across modes/tasks Model-agnostic meta-learning (MAML) [1] • Seek a common initialization parameter for all the modes An ensemble of MAMLs (Multi-MAML) • Mode labels are often not available • Prevent sharing related knowledge among modes/tasks
![Page 1: via Task-Aware Modulation - GitHub Pages• Task network: further gradient adaptation via MAML steps Background Model-Agnostic Meta-Learning [1] • Meta-learn a parameter initialization](https://reader036.documents.pub/reader036/viewer/2022062506/5f01ee407e708231d401bd1c/html5/thumbnails/1.jpg)
![Reconciling meta-learning and continual learning with online … · In particular, [21] cast MAML as posterior inference for task-specific parameters j given some samples of task-specific](https://static.documents.pub/doc/80x56/6119a446d76d8963921e1328/reconciling-meta-learning-and-continual-learning-with-online-in-particular-21.jpg)
![arXiv:2002.04766v1 [cs.LG] 12 Feb 2020Distribution-Agnostic Model-Agnostic Meta-Learning Liam Collins , Aryan Mokhtari , Sanjay Shakkottai February 13, 2020 Abstract TheModel-AgnosticMeta-Learning(MAML)algorithm[Finnetal.,2017]hasbeencel](https://static.documents.pub/doc/80x56/5e5f5b26282d4a2232338b23/arxiv200204766v1-cslg-12-feb-2020-distribution-agnostic-model-agnostic-meta-learning.jpg)
![Visual Question Answering as a Meta Learning Task...Visual QuestionAnswering as a Meta Learning Task Damien Teney[0000−0003−2130−6650], Anton van den Hengel[0000−0003−3027−8364]](https://static.documents.pub/doc/80x56/60cfcfd03a70c467b457f70a/visual-question-answering-as-a-meta-learning-task-visual-questionanswering-as.jpg)
![Adaptive Gradient-Based Meta-Learning Methods · GBML: GBML stems from the Model-Agnostic Meta-Learning (MAML) algorithm [23] and has been widely used in practice [1, 44, 31]. An](https://static.documents.pub/doc/80x56/5fe123ce63e4bc70da1aaedb/adaptive-gradient-based-meta-learning-methods-gbml-gbml-stems-from-the-model-agnostic.jpg)
