Webly Supervised Semantic Embeddings for Large Scale Zero-Shot Learning

Cacheux, Yannick Le; Popescu, Adrian; Borgne, Hervé Le

doi:10.1007/978-3-030-69544-6_31

Cited by 8 publications

(3 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We adopt a linear ridge regression model, in which semantic prototypes are projected into the visual features space so as to minimize a regularized least square loss. This choice is motivated by the good and consistent results of this model in [14], as well as its low computational cost and ease of reproducibility. The choice of the visual space as the projection space is based on [25], which argues that this choice enables to mitigate the hubness problem [23].…”

Section: Proposed Methodsmentioning

confidence: 99%

“…Glove embeddings are employed in [8]; we also compare with Word2vec and FastText [2], as well as Elmo contextual embeddings using pre-trained embeddings available on the Internet (details are provided in the supplementary material). The ZSL model is a ridge regression from the semantic to the visual space as it gives good, consistent and easily reproducible results (see [14]).…”

Section: Visualness Based Approachmentioning

confidence: 99%

“…Word embeddings are then typically used to represent classes. However, a large performance gap still exists between these two types of class representations [14].…”

Section: Introduction and Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Using Sentences as Semantic Representations in Large Scale Zero-Shot Learning

Cacheux,

Borgne,

Crucianu

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Zero-shot learning aims to recognize instances of unseen classes, for which no visual instance is available during training, by learning multimodal relations between samples from seen classes and corresponding class semantic representations. These class representations usually consist of either attributes, which do not scale well to large datasets, or word embeddings, which lead to poorer performance. A good trade-off could be to employ short sentences in natural language as class descriptions. We explore different solutions to use such short descriptions in a ZSL setting and show that while simple methods cannot achieve very good results with sentences alone, a combination of usual word embeddings and sentences can significantly outperform current state-of-the-art 3 .

show abstract

Section: Proposed Methodsmentioning

confidence: 99%