Frameworks for deep learning with dynamic graphs construction

Two frameworks PyTorch and DyNet are released recently. They both feature dynamic graph construction, which is more flexible than current frameworks like Tensorflow and Theano, with other features of their own. (though Tensorflow has some functions for dynamic NN architecture construction)

DyNet: https://arxiv.org/abs/1701.03980

PyTorch: http://pytorch.org/


Update on Feb 16:

Tensorflow accounced a new feature called Tensorflow Fold, which supports dynamic computation graphs. More importantly, different from the above two, it also supports dynamic batching, which makes it more powerful. I would expect other frameworks like MXNet/Pytorch/others will support the function as well very soon.

Details here:

 

More to be continued on these topics soon.

 

Review of Neural Machine Translation

This is a post that reviews the recent advancements of machine translation, especially neural machine translation (NMT). 

Prerequisites: this post assumes some prior knowledge about machine learning, artificial neural networks, CNN, RNN (LSTM, GRU) encoder-decoder architecture, seq-to-seq models, etc. Continue reading