RNN sequence labeling for document parsing in Tensorflow
summary
CV parsing uses Tensorflow to train a BLSTM sequence labeling model that cuts errors by about 30% error reduction compared with CRF/HMM baselines. Implementation details include word embeddings, bidirectional LSTM stacks, optional CRF layer for transitions, and training on a few thousand annotated resumes. Model export works via SavedModel and simple Flask or tf Serving endpoints, training benefits from cuDNN LSTM on GPU and TensorBoard visualisation for debugging.
AI-written summary · about
chapters (12)
- 0:00Document parsing with RNNs
- 4:52Traditional to deep learning
- 7:19LSTM sequence labeling
- 9:48Setting up TensorFlow for sequence labeling
- 13:37Bidirectional LSTM network
- 18:27Predictions with dense layer
- 20:25Training with masked loss
- 25:59Training and visualization
- 29:41Debugging and evaluation
- 31:15Production deployment
- 32:17Challenges and tuning
- 35:05Q&A
related talks
53mTensorflow, deep learning and modern RNN architectures, without a PhD
19mDocument Reader using Deep Learning with a Small Dataset
40mSection Extraction from Texts
19mRushin Shah, Engineering Manager, Facebook
30mStarting deep learning projects without enough labelled data
22mHow to expand your NLP Solution to new languages
01h 26mFrom RNNs to Transformers: A Journey through Language Modeling
54mNatural Language Processing with Deep Learning and TensorFlow