Pointer Networks
A neural network architecture proposed by Vinyals et al. in 2015 that uses an attention mechanism to directly point to positions in the input sequence as output. The output vocabulary size is dynamically bound to the input length, so outputs necessarily belong to the input set, making it naturally suited to combinatorial optimization, extractive question answering, and entity recognition.
Timeline (last 90 days)
Pointer Networks were proposed by Vinyals et al. in 2015 and are a neural network structure that uses an attention mechanism to point directly at a position in the input sequence.
The output vocabulary size of a pointer network is dynamically tied to the input length, and each output token corresponds to an index in the input rather than being generated from a fixed vocabulary.
All Facts (2)
Pointer Networks were proposed by Vinyals et al. in 2015 and are a neural network structure that uses an attention mechanism to point directly at a position in the input sequence.
50%UnverifiedThe output vocabulary size of a pointer network is dynamically tied to the input length, and each output token corresponds to an index in the input rather than being generated from a fixed vocabulary.
50%