Title:
|
Skip RNN: learning to skip state updates in recurrent neural networks
|
Author:
|
Campos Camunez, Victor; Jou, Brendan; Giró Nieto, Xavier; Torres Viñals, Jordi; Chang, Shih-Fu
|
Other authors:
|
Universitat Politècnica de Catalunya. Departament de Teoria del Senyal i Comunicacions; Universitat Politècnica de Catalunya. Departament d'Arquitectura de Computadors; Universitat Politècnica de Catalunya. GPI - Grup de Processament d'Imatge i Vídeo; Universitat Politècnica de Catalunya. CAP - Grup de Computació d'Altes Prestacions |
Abstract:
|
Recurrent Neural Networks (RNNs) continue to show outstanding performance in sequence modeling tasks. However, training RNNs on long sequences often face challenges like slow inference, vanishing gradients and difficulty in capturing long term dependencies. In backpropagation through time settings, these issues are tightly coupled with the large, sequential computational graph resulting from unfolding the RNN in time. We introduce the Skip RNN model which extends existing RNN models by learning to skip state updates and shortens the effective size of the computational graph. This model can also be encouraged to perform fewer state updates through a budget constraint. We evaluate the proposed model on various tasks and show how it can reduce the number of required RNN updates while preserving, and sometimes even improving, the performance of the baseline RNN models. Source code is publicly available at https://imatge-upc.github.io/skiprnn-2017-telecombcn/ |
Subject(s):
|
-Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Llenguatge natural -Neural networks (Computer science) -Graph theory -Natural language processing (Computer science) -Knowledge representation (Information theory) -recurrent neural networks -dynamic learning -conditional computation -Xarxes neuronals (Informàtica) -Tractament del llenguatge natural (Informàtica) -Grafs, Teoria de -Representació del coneixement (Teoria de la informació) |
Rights:
|
|
Document type:
|
Article - Published version Conference Object |
Share:
|
|