Machine Learning & Neural Networks Demystified
summary
A trivago engineering team explains an image-processing pipeline implemented in Python that trains convolutional neural networks for object detection using GPU-parallelized matrix operations. They describe the full workflow—convolution and pooling feature maps, backpropagation/gradient descent—and give concrete data (e.g., MNIST 70,000, 784→128 hidden→~100k weights) and real-world practices like label sourcing, external annotators, segmentation experiments, and transfer-learning approaches to justify their design choices.
AI-written summary · about
chapters (14)
- 0:00Neural networks intro
- 1:29Python for ML
- 2:22How brain decides
- 4:14Neural networks basics
- 6:39Math behind networks
- 9:11Cost function & error
- 11:43Gradient descent
- 15:08Training steps
- 18:41Matrix math for nets
- 19:44MNIST example
- 21:10Simple nets to CNNs
- 22:36How CNNs recognize images
- 25:55Deep learning resources
- 27:15Q&A
