文章主要问题是解决少样本学习,灵感来自actor-critic增强学习,但可以应用于增强和监督学习。核心方法是学习一个meta-critic——神经网络的行为价值函数,学习去评判解决特殊任务的actor。对于监督学习,相当于一个可训练的任务参数损失发生器。对于增强学习和监督学习,这种方法提供了一种知识迁移途径,可以处理少样本和半监督条件。
相关文章
- 03-23【Meta learning】Learning to learn: Meta-Critic Networks for sample efficient learning
- 03-23Learning bothWeights and Connections for Efficient Neural Networks 论文阅读
- 03-23Learning Efficient Convolutional Networks Through Network Slimming论文阅读笔记
- 03-23(转)Paper list of Meta Learning/ Learning to Learn/ One Shot Learning/ Lifelong Learning
- 03-23【论文考古】联邦学习开山之作 Communication-Efficient Learning of Deep Networks from Decentralized Data
- 03-23Sample-Efficient Reinforcement Learning with Stochastic Ensemble Value Expansion
- 03-23《Learning to warm up cold Item Embeddings for Cold-start Recommendation with Meta Scaling and Shifting Networks》论文阅读
- 03-23模型压缩之Learning Efficient Convolutional Networks through Network Slimming
- 03-23meta learning - Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks - 1 - 论文学习
- 03-23论文笔记:Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks