Dynamic Neural Networks using PyTorch
summary
Introduction to building deep learning models with PyTorch, focusing on dynamic neural networks whose structure adapts per input and on using standard programming abstractions (functions, objects) and debugging tools. Explains how PyTorch and similar imperative frameworks (Gluon) differ from Torch, Keras, and TensorFlow by using auto-differentiation to let developers define complex optimization objectives without manual derivative or CUDA coding, and reviews best practices and production considerations for PyTorch.
AI-written summary · about
chapters (20)
- 0:00PyTorch introduction
- 3:06Static vs dynamic neural networks
- 7:19Computational graphs paradigm
- 9:05Dynamic PyTorch advantages
- 10:31PyTorch history and evolution
- 12:38Tensors data structure
- 14:03Automatic differentiation
- 19:46PyTorch basic blocks
- 21:17Data preprocessing
- 23:45Building dynamic models
- 27:21Optimizers and training
- 29:15Saving and loading models
- 30:42Accessing internal representations
- 33:14Image classification with PyTorch
- 37:07Convolutional network architecture
- 39:18Evaluating trained models
- 42:32GPU and CPU usage
- 44:01Stability and limitations
- 45:06Future PyTorch improvements
- 46:15Closing and Q&A
related talks
41mReally Deep Neural Networks with PyTorch
48mDynamic Neural Networks with DyNet
43mExploring Deep Learning Framework PyTorch
01h 26mDeep Neural Networks with PyTorch
47mGet to Know PyTorch - A Cutting Edge Open Source AI Framework for Deep Learning
03h 00mThe Fundamentals of Modern Deep Learning with PyTorch
01h 05mDeep Learning with PyTorch
36mPyTorch: Fast Differentiable Dynamic Graphs in Python