variational-siamese-network
Overview
Tensorflow implementation of Learning semantic similarity in a continuous space
Requirements
Usage
Pretrained word-2-vec language model, under w2v, from GloVe or Word2Vec.
Semantic Similarity Dataset from quora_duplicate_questions.tsv for train/dev/test, saved under data/split folder as text files.
- To train a variational siamese network with generative pretraining (repeat), run the following snipped in the folder _deepNLU:
> python train.py VAE
- For the repeat, reformulate framework, run:
> python train.py VAD
- To visualize training on tensorboard, run:
> tensorboard --logdir=summary
- To test a trained model / for inference, run eval.ipynb
Architecture and Training
Variational Auto Encoder (VAE)
Variational Siamese Network
Semantic Similarity Evaluation
Acknowledgments
Ecole Polytechnique,Télécom Paris-Tech
Pr. Chloé Clavel, Pr. Gabriel Peyré, Pr. Francis Bach, Pr. Guillaume Obozinski, Pr. Michalis Vazirgiannis
Magdalena Fuentes, Constance Nozière and Paul Bertin
Reviewers for their valuable comments and feedback