Baby steps towards few-shot learning with multiple semantics

Learning from one or few visual examples is of the key capabilities humans since early infancy, but still a significant challenge for modern AI systems. While considerable progress has been achieved in few-shot learning image examples, much less attention given to verbal descriptions that are usually provided infants when they presented with new object. In this paper, we focus on role additional semantics can significantly facilitate learning. Building upon recent advances semantic information, demonstrate further improvements possible by combining multiple and richer (category labels, attributes, natural language descriptions). Using these ideas, offer community results popular miniImageNet CUB benchmarks, comparing favorably previous state-of-the-art both only plus semantics-based approaches. We also performed an ablation study investigating components design choices our approach. Code available github.com/EliSchwartz/mutiple-semantics.