Sequential Attention-Based Neural Machine Translation
summary
Shows a working neural machine translation built implementation from scratch using TensorFlow, training on an English↔Portuguese dataset (prayer sentences) and the GitHub code. Explains the exact attention math: attention masks and multi-head attention, query/key/value dot products, layer norm and dropout. Demonstrates training details (batching, tokenization, epochs) and claims useful output after only minutes of training — attention heatmaps and sample translations are produced so viewers can verify whether the model actually works.
AI-written summary · about
chapters (10)
- 0:00Neural machine translation intro
- 5:35Recurrent neural networks background
- 12:41Data prep for translation
- 16:53Challenges in sequence-to-sequence translation
- 21:57Masking and attention
- 28:41Transformer core components
- 35:22Multi-head attention
- 37:50Encoder and decoder layers
- 40:44Training the model
- 45:01Model evaluation and results
related talks
01h 25mPaper Discussion: Attention is all you need
42mAttention and Memory in Deep Learning Networks
01h 29mBuild Your Own Transformer
24mRevealing the Unseen: Visualizing Attention in Transformer Models
26mUsing deep learning in natural language processing
28mEverything you need to know about Transformer Models
36mNeural Networks
39mDeep Learning for Machine Translation: a paradigm shift