Introduction to Deep Learning & Natural Language Processing
summary
Repository is cloned and notebooks run; hands‑on includes MLP coded from scratch and a Keras Sequential MLP on a Kaggle auto dataset (input dim 93, dense 64 → softmax) that shows ~0.64 accuracy after the first epoch and improves with more epochs (batch 128). A gensim/word2vec demo on Gutenberg plays (Julius Caesar + Macbeth) uses window=5 and high‑dim embeddings with t‑SNE analogies. A char‑level LSTM trains on ~3,000 blogs for gender classification (epochs ~10, ~190s/epoch on CPU). Notes on dropout, SGD/mini‑batches, GPU speedups and the recommendation that RNNs need much larger corpora (hundreds of thousands–millions) to generalize.
AI-written summary · about
chapters (26)
- 0:00Talk introduction
- 1:14Speaker introduction
- 2:18What is machine learning
- 3:45Biological inspiration
- 5:53How neurons work
- 8:18Visual cortex learning
- 10:07Linear separation challenges
- 12:12Neural network building blocks
- 13:43Perceptrons and activation
- 17:19Backpropagation learning
- 21:48Gradient descent techniques
- 24:51Multi-layer perceptron intro
- 27:26MLP with Keras demo
- 30:17Environment setup demo
- 37:35Overfitting and dropout
- 42:24Natural Language Processing intro
- 44:25Traditional NLP challenges
- 49:01Vectorization and word rep
- 51:52Word2Vec applications
- 1:02:22Word2Vec on Shakespeare
- 1:09:48Recurrent Neural Networks and their types
- 1:12:38LSTM networks explained
- 1:17:26LSTM author classification
- 1:21:57Hardware acceleration
- 1:25:19Real-world deep learning apps
- 1:28:56Q&A
related talks
31mDeep Learning Techniques for Natural Language Processing
01h 59mDeep Learning and Modern NLP
01h 40mA Deep Dive into NLP with PyTorch
45mDeep Learning for Natural Language Processing
50mDeep Learning and NLP with Spark
31mRemaining Challenges in Deep Learning based NLP
35mDeconstructing Deep Learning
01h 32mNeural Networks for Natural Language Processing