A practical guide to deep learning
summary
A practical pipeline builds a Chihuahua vs muffin classifier from web images using Keras: roughly 500 images per class, train/validation/test splits, on-batch training with data augmentation and dropout to fight overfitting, and runtime around 16s per epoch on a GTX1060. Accuracy jumps to ~99.47% when using transfer learning with VGG16 pretrained on ImageNet. Model inspection uses LIME local interpretability to find spurious background features; warnings include biased training datasets and metrics that mask skewed classes.
AI-written summary · about
chapters (3)
related talks
01h 01mA practical guide to deep learning
32mYou Can Do Deep Learning!
02h 26mPractical Deep Learning for Data Scientists
01h 00mA gentle introduction to Deep Learning
40mHow to Implement a Production Ready Deep Learning Model in 12 Weeks
35mDeconstructing Deep Learning
51mA Practical Introduction to Convnets for Image Classification
53mDeep Learning in the world of little ponies