Swifty TensorFlow
summary
Argues that Swift for TensorFlow turns eager-style coding into deployable graphs via Graph Program Extraction, exposes gradients by letting the compiler generate chain-rule derivatives and offers automatic differentiation built directly into Swift. Demonstrates a Swift model training walkthrough running in Colab with Python interop, shows debugging and stepping through compiled graphs, and contrasts Swift for TensorFlow 0.3 with TensorFlow 2.0 eager execution as the practical stack that actually lets developers inspect and iterate on models.
AI-written summary · about
chapters (25)
- 0:00Intro
- 0:54Why Machine Learning
- 5:20What is Machine Learning
- 5:56Star Trek vs AI
- 9:29Feature Engineering / Synthetic Features
- 11:02Setup the problem
- 11:32Encode the data
- 14:51Reducing Loss with Supervised Learning
- 18:03Layered Neural Network
- 18:57Some Machine Learning Terms ...
- 22:32Logistic Regression: Calculating a Probability
- 25:22Training Error vs Testing Error
- 25:56Validation Set / Test Set
- 27:13Dropout (Rate)
- 27:45Batch Normalisation
- 29:10Gradient Descent on Solution Space (W & b)
- 30:45Training with ... Partial Differential Equations
- 31:44Automatic Differentiation
- 35:12Hyper Parameters
- 36:14CNN Constructing Features
- 36:46Machine Learning for Images
- 39:22Build a Convolutional Neural Network
- 40:23Prototyping Platforms
- 43:01Differentiable Programming
- 43:13Building the Eger Execution
related talks
28mSwift for TensorFlow
40mIntroduction to Swift for Tensorflow
39mSwift for TensorFlow: Machine Learning with No Boundaries
24mSwift for TensorFlow: machine learning with no boundaries.
32mGetting Started with TensorFlow 2.0
21mWhy TensorFlow eager execution matters
41mTensorFlow ever higher
41mLearning TensorFlow