Label Representations • Let l ∈ Rd be the representation of label l, and f be a differentiable embedding function • For labels of label type i, we apply a learnable embedding function l = fi(l) • hi(v) is the embedding of label type i for vertex v: hi(v) = gi ({l | l ∈ labels of type i associated with vertex v}) • h̃i(v) is the reconstruction of the embedding of label type i for vertex v: h̃i(v) ...

We propose a method, called Label Embedding Network, which can learn label representation (label embedding) during the training process of deep networks. With the proposed method, the label embedding is adaptively and automatically learned through back propagation. The original one-hot represented loss function is converted into a new loss function with soft distributions, such that the origina...

Zero Shot Learning (ZSL) enables a learning model to classify instances of an unseen class during training. While most research in ZSL focuses on single-label classification, few studies have been done in multi-label ZSL, where an instance is associated with a set of labels simultaneously, due to the difficulty in modeling complex semantics conveyed by a set of labels. In this paper, we propose...

The standard approach to recognizing text in images consists in first classifying local image regions into candidate characters and then combining them with high-level word models such as conditional random fields (CRF). This paper explores a new paradigm that departs from this bottom-up view. In our approach, every label from a lexicon is embedded to an Euclidean vector space. We refer to this...

Automatically tagging textual mentions with the concepts, types and entities that they represent are important tasks for which supervised learning has been found to be very effective. In this paper, we consider the problem of exploiting multiple sources of training data with variant ontologies. We present a new transfer learning approach based on embedding multiple label sets in a shared space,...