Key Points

Introduction to Deep Learning

Deep learning uses neural networks to learn patterns directly from data.
Convolutional neural networks (CNNs) are commonly used for image classification.
Training a model involves compiling it, fitting it to data, and making predictions.
Model performance may be imperfect at first and can be improved with further training and tuning.

Images consist of pixels arranged in a particular order.
Image datasets can be organised into folders where each folder represents a class.
image_dataset_from_directory() lets us load images without writing custom data preparation code.
Images are loaded in batches and represented as numerical arrays.
Training, validation, and test sets are used to build and evaluate models.

Convolutional neural networks (CNNs) are designed for working with image data.
CNNs are built by stacking layers, where each layer transforms the data and passes it to the next layer.
Convolutional layers look for simple patterns in images (e.g. edges and textures).
Pooling layers reduce the size of the data, helping the model focus on important features.
The Flatten layer converts image data into a format suitable for classification.
Dense layers are used to produce the final prediction.

Use Model.compile() to set how a model will learn.
The optimizer controls how the model updates its weights.
The loss function measures how wrong the model’s predictions are.
Metrics such as accuracy tell us how well the model is performing.
Use Model.fit() to train the model on data.
Training and validation loss and accuracy help us monitor learning.
Overfitting occurs when a model performs well on training data but less well on new data.

Use Model.predict() to make predictions with a trained model.
Model outputs represent confidence values for each class.
Use argmax to convert model outputs into predicted class labels.
Accuracy measures how often predictions are correct.
Confusion matrices help identify which classes are predicted correctly and where errors occur.