This presentation provides a theoretical basis for convolutional neural networks. It also offers an operational approach to solving problems that require image analysis. I suggest to follow these steps:
- Construction and training of a standard convolutional neural network using the code provided on the Jupiter notebook
- Identification of problems (overfitting, underfitting, long training time) starting from the analysis of the loss function
- Modification of the hyperparameters according to the indications provided in the last paragraph of the presentation
- Model adaptation and performance improvement
I built the whole presentation with Prezi. This tool is fascinating because it is based on studies on the graphic formats that the human brain prefers to store information better. Prezi allows in a short time to create presentations that are different from the average, clear and impressive.