Deep Learning for Natural Language Processing

on YouTube ↗45m

" summary

Demonstrates using CNN with word embeddings to classify short texts, replacing heavy feature engineering and achieving large gains: reported jump from 85% to 95% accuracy on a binary sentiment set (73% error reduction). Uses Google News embeddings, Theano implementations, GPU training (30× speedup; Flipkart experiment trains ~1500/500 split in ~2 hours on a single GPU laptop). Releases code as the Optimus toolkit on GitHub and shows which window sizes and pooling choices produced the surviving features.

AI-written summary · about

" chapters (18)

1. 0:00Start
2. 1:08THE NEED FOR NATURAL LANGUAGE PROCESSING
3. 7:29IMAGE MODELS
4. 7:57DEEP LEARNING FOR NLP
5. 9:37STATISTICAL CLASSIFIERS
6. 10:30FEATURE ENGINEERING
7. 16:15PROBLEMS WITH STATISTICAL MODELS
8. 17:46WORD EMBEDDINGS: THE FIRST BREAKTHROUGH
9. 18:15NEURAL LANGUAGE MODEL
10. 19:00WORD EMBEDDINGS: VISUALIZATIONS
11. 19:22CAPTURE RELATIONSHIPS
12. 20:24WORD EMBEDDING: VISUALIZATIONS
13. 21:07COMPOSITIONALITY
14. 22:13CONVOLUTIONAL NEURAL NETWORKS
15. 25:53DEMYSTIFYING MAX POOLING
16. 28:42SENTIMENT: ANECDOTES
17. 29:45DRAWBACKS & LEARNINGS
18. 32:19RECURRENT MODELS

" related talks

• 01h 32mNeural Networks for Natural Language ProcessingMatti Lyra · PyData New York City 2019
• 31mDeep Learning Techniques for Natural Language ProcessingParis Buttfield-Addison · Developer Summit 2020
• 01h 30mIntroduction to Deep Learning & Natural Language ProcessingRaghotham Sripadraj, Nischal HP · PyData London 2016
• 35mDeconstructing Deep LearningMark West · Big Data Conference Europe 2019
• 34mWord Embeddings for fun and profit in GensimLev Konstantinovskiy · PyData London 2016
• 43mDemocratizing NLP content modeling with transfer learning using GPUsSanghamitra Deb · PyData DC 2018
• 39mDeep Learning vs. Conventional Machine LearningChao Han · PyData Miami 2019
• 54mNatural Language Processing with Deep Learning and TensorFlowBarbara Fusinska · NDC Porto 2019