Lessons learned from the NeurIPS 2021 MetaDL challenge: Backbone fine-tuning without episodic meta-learning dominates for few-shot learning image classification.
Adrian El BazIhsan UllahEdesio AlcobaçaAndré C. P. L. F. de CarvalhoHong ChenFabio FerreiraHenry GoukChaoyu GuanIsabelle GuyonTimothy M. HospedalesShell HuMike HuismanFrank HutterZhengying LiuFelix MohrEkrem ÖztürkJan N. van RijnHaozhe SunXin WangWenwu ZhuPublished in: NeurIPS (Competition and Demos) (2021)
Keyphrases
- lessons learned
- meta learning
- learning tasks
- fine tuning
- image classification
- meta knowledge
- inductive learning
- learning algorithm
- case study
- learning process
- decision trees
- feature extraction
- prior knowledge
- model selection
- supervised learning
- machine learning algorithms
- knowledge acquisition
- data sets
- software engineering
- machine learning
- data mining