Deep Learning in the world of little ponies
summary
Builds a six-class My Little Pony image classifier from about 400 training and 100 test images using Keras (Python) and shows an alternative quick route with Azure Custom Vision (custom model trained from ~80 images per class, then exported to run on-device). Demonstrates a CNN with 150×150 inputs, data augmentation, dropout, ten epochs and batch size ~30; MNIST baseline hits ~97.7% while the pony model is inconsistent (some predictions 0.99, others misclassified due to background), then applies transfer learning and exports a compact model for mobile/edge export.
AI-written summary · about
chapters (33)
- 0:00Intro
- 0:23Who is Galiya
- 1:51Finding data
- 3:03Cognitive services
- 3:51Custom Vision AI
- 5:37Adding images
- 7:34Training
- 8:08Testing
- 10:20How to use it
- 13:33What is an image
- 14:25Before Deep Learning
- 16:50Feature Extraction
- 19:00Representations
- 20:28Neurons
- 21:24Layers of Neural Network
- 21:58Activation Functions
- 24:28Karis
- 25:43Traditional neural network architectures
- 26:50Neural network layers
- 28:05Databricks
- 33:08Image Classification
- 39:56Visualization
- 42:33Convolution Layer
- 45:01Pooling Layer
- 46:06Feature Map
- 46:32Fully Connected Layer
- 47:39Overfitting
- 48:50Date Augmentation
- 49:41Dropout
- 50:28Demo
- 56:40Transfer Learning
- 57:55Demonstration
- 59:53Steel Model
