[Adapt] [Seminar] Few Shot Learning and Meta Learning

[耿晓晴]

Hi Adapters,

I will share with you something about Few Shot Learning and Meta Learning this week.

Deep learning requires large amount of data to train a neural networks with good performance. Given less data, deep learning may not achieve as good performance. Meta learning deals with the problem of learning to learn. In meta learning, we create neural networks with a lot of background knowledge and make it adjust quickly to a new tasks without needing a ton of data. In my talk, I will introduce the motivation and definition of meta learning, and two related papers: 
1. One-Shot Learning with Memory-Augmented Neural Networks (Santoro et al. 2016), which works on few-shot image classification task
2. FewRel: A Large-Scale Supervised Few-Shot Relation Classification Dataset with State-of-the-Art Evaluation (Xu et al. 2018, EMNLP), which works on few-shot relation classification task

See you then!

Time: 17:00 April 10
Venue: SEIEE 3-414

Best,
Xiaoqing