Языковая модель на основе рекуррентных нейронных сетей

We investigate an extension of continuous online learning in recurrent neural network language models. The model keeps a sep- arate vector representation of the current unit of text being processed …...

We introduce a recurrent neural network lan- guage model (RNN-LM) with long short- term memory (LSTM) units that utilizes both character-level and word-level inputs. Our model has a gate that …...

AbstractWe propose algorithms to train production-quality n-gram language models using federated learning. Federated learning is a distributed computation platform that can be used to train global models for portable devices …...