Project summary

The core principle of few-shot learning is to build on visual feature representations obtained by pre-training neural networks on large image datasets and transfer the learned representation to recognize new object classes from a few samples. We hypothesize that considerable gains can be made by investigating and expanding the foundations of few-shot learning. We aim to provide a concise and solid framework that generalizes to diverse applications. Specifically, we will develop novel core algorithms for few-shot learning, including analysis of the properties of base representations, matching strategies, generative techniques to learn residual features for distinguishing new objects, and techniques for incrementally learning new classes without forgetting previous knowledge.

Our methods

• Data-driven modeling
• Neural networks
• Meta-learning
• Pre-training
• Local perspective