목록meta-learning (6)

iMTE

Meta-learning with Implicit Gradients (NIPS 2019) [1]

Meta-learning with Implicit Gradients [1] https://papers.nips.cc/paper/2019/hash/072b030ba126b2f4b2374f342be9ed44-Abstract.html IntroductionMeta-learning의 frame에서 bi-level optimization procedure는 다음으로 나누어진다.1) inner optimization : 주어진 task에 base learner가 학습하는 과정2) outer optimization : 여러 tasks 들에서 meta learner가 학습하는 과정MAML, DAML, Reptile 등의 방법이 optimization-based methods에 속한다. (Hands-on one-shot..
Deep learning study/Meta-Learning 2020. 12. 24. 15:23
Meat-Learning 8. Meta-learning for semi-supervised few-shot classification

Meta-learning for semi-supervised few-shot classification (ICLR, 2018) Abstract"In this work, we advance this few-shot classification paradigm towards a scenario where unlabeled examples are also available within each episode....To address this paradigm, we propose novel extensions of Prototypical Networks that are augmented with the ability to use unlabeled examples when producing prototypes."U..
Deep learning study/Meta-Learning 2019. 5. 25. 16:13
Meta-Learning 7. Deep Meta-Learning: Learning to Learn in the Concept Space

Deep Meta-Learning: Learning to Learn in the Concept Space Abstract "In this work, we argue that this is due to the lack of a good representation for meta-learning, and propose deep meta-learning to integrate the representation power of deep learning into meta-learning" Few-shot learning이 많은 related tasks로 부터 learning algorithm을 학습하는 것이 어려운데, 이를 좀 더 나은 representation으로 해결하자는 방법이 Deep Meta-Learni..
Deep learning study/Meta-Learning 2019. 5. 24. 16:19
Meta-Learning 6. Gradient agreement as an optimization objective for meta-learning

Gradient agreement as an optimization objective for meta-learning (NIPS, 2018) Abstract"Our approach is based on pushing the parameters of the model to a direction in which tasks have more agreement upon. If the gradients of a task agree with the parameters update vector, then their inner product will be a large positive value." 각 Task에서 얻은 gradient의 정보에서 가장 agreement가 큰 부분을 살리고, 반대되는 경우나 다른 방향의..
Deep learning study/Meta-Learning 2019. 5. 24. 14:54
Meta-Learning 5. ADML, Meta-SGD, and Reptile

Meta-Learning sub parts 1. Adversarial Meta Learning (ADML)MAML에서 optimal weight initialization을 찾는 것이 목표였다. Optimal weight initialization을 찾을 때, 좀더 나으면서 견고한 model parameters를 찾기 위해서 clean data와 adversarial data를 사용하는 MAML을 ADML이라고 부른다. Network가 adversarial examples에 의해서 잘못된 판단을 내린다는 점으로 인해서, adversarial example을 사용하는 경우 좀 더 network가 adversarial attack에 강인하고, noisy에 강인하다는 점을 갖게 된다. 이러한 성질을 MAML에..
Deep learning study/Meta-Learning 2019. 5. 18. 20:16
Meta-Learning 1. Learning to learn by gradient descent by gradient descent

Paper : Learning to learn by gradient descent by gradient descentDownload : http://papers.nips.cc/paper/6460-learning-to-learn-by-gradient-descent-by-gradient-descentMeta-Learning을 공부하고 연구에 적용해볼 생각으로 정리해보려고 한다. 꽤 재밌는 아이디어고, 어느 tasks에 적용할 수 있다는 점에서 매우 강력한 알고리즘이 될 것이라고 생각한다. Abstract기존의 hand-crafted features 대신에 학습된 features로의 변화는 꽤 큰 성공이었다. 하지만, 최적화 알고리즘 (e.g., optimization algorithm)은 여전히 hand로 ..
Deep learning study/Meta-Learning 2019. 4. 20. 14:34
Prev 1 Next