Notice

Machine Intelligence and Information Learning (MIIL) lab’s paper titled “XtarNet: Learning to Extract Task-Adaptive Representation for Incremental Few-Shot Learning” is accepted to International Conference on Machine Learning (ICML), which is one of top conferences for artificial intelligence (acceptance rate 21.8%).

Learning novel concepts while preserving prior knowledge is a long-standing challenge in machine learning. The challenge gets greater when a novel task is given with only a few labeled examples, a problem known as incremental few-shot learning.  In this paper, Prof. Sung Whan Yoon and co-authors propose a novel meta-learning algorithm called XtarNet, which can learn new concepts with only a few labeled samples while preventing catastrophic forgetting of previous knowledge. The proposed algorithm can be a proper solution for resolving the data-dependency of current deep-learning methods, and also extended to continual/lifelong learning which are very important research directions for future intelligent systems.

Authors: Sung Whan Yoon*, Do-Yeon Kim*, Jun Seo and Jaekyun Moon

*Equal contribution; Mr. Kim, Mr. Seo and Prof. Moon are at KAIST.

Paper link: https://arxiv.org/abs/2003.08561