Preview
IB DP Computer Science — A4 Machine learning — Machine learning approaches | HL | A4.3.9
📋 What You Will Learn
- Outline the basic CNN architecture: input, convolutional layers, activation functions, pooling layers, fully connected layers, output
- Compute a convolution and a max-pooling operation by hand
- Describe how stacked layers build a spatial hierarchy: edges → parts → objects
- Explain the effects of the number of layers, kernel size, stride, activation function and loss function
- Explain why "adaptively learn" means the filters themselves are trained, not hand-designed
👁️ Why This Matters You never recognise a friend by checking every pixel of their face independently. Your visual system first finds edges, then combines edges into features like eyes and noses, then combines those into a face. A convolutional neural network is engineered around exactly that idea — and it is why computers can now read scans, drive cars and sort photos.
Why Images Break Ordinary ANNs
- Flattening even a small 200×200 colour image into an MLP input gives 120 000 numbers; a first hidden layer of 1 000 neurons already needs 120 million weights — slow to train and