ML from a software engineer's perspective
summary
Shows how to build a digit classifier using TensorFlow high-level API applied to an MNIST-style dataset, trains a small neural network on CPU in about 15 minutes, and reports solid test accuracy while demonstrating overfitting when hidden layers or neurons grow. Visualizations map pixels into a feature space to explain decision boundaries, training steps, and when to stop. Emphasizes starting small, iterating on feature choice, and using validation/test sets to decide improvements.
AI-written summary · about
chapters (12)
- 0:00Introduction to ML for developers
- 3:25What is machine learning
- 5:11Neural networks crash course
- 8:47High-level APIs benefits
- 10:17Feature selection decisions
- 13:14Visualizing neural network learning
- 18:37Overfitting and test sets
- 21:48Network complexity intuition
- 23:00Implementing digit recognition
- 27:25First results and next steps
- 30:21Resources and further learning
- 33:54Q&A and closing
related talks
01h 07mDeep Learning with TensorFlow
45mThe most soft and fluffy intro into Machine Learning and Deep Neural Networks
01h 06mDeep Learning with TensorFlow
42mStep into Machine Learning .NET
51mMachine Learning Explained to Humans
50mLearning Machine Learning
43mA Gentle Introduction to Neural Networks
45mML: ML.NET: deep dive into dev AI