Meta-Learning: Learning to Learn Fast
Lilian Weng
Meta-learning is a machine learning approach designed to enable models to learn new concepts and tasks quickly from few examples, similar to how humans learn. The method trains models on a distribution of tasks by optimizing performance across multiple datasets, with techniques including metric-based approaches using siamese networks and matching networks that learn similarity functions. This approach enables models to adapt to new tasks during testing with limited data exposure, contrasting with standard supervised learning that requires large numbers of training samples.
Why it matters
[Updated on 2019-10-01: thanks to Tianhao, we have this post translated in Chinese!]
