Machine Learning projects using classification, neural networs and CNNs
Mestrado Engenharia Informática e Multimédia - ISEL
The first project is based on the binary and multi-class classification of images using the CIFAR-10 database.
- In both tasks it was verified if the usage of normalization was beneficial.
- In the multi-class classification task, the data was transformed with PCA.
- The results were analyzed with different methods like confusion matrixes and score metrics.
The second project is based on implementing, training and testing Multi-Layer Perceptron (MLP) neural networks.
-
A three-layer MLP network, for the task of binary classification of two-dimensional data (XOR pro-blem) was implemented without resorting to any library like TensorFlow/Keras.
-
In the second objective several MLP networks were trained with the databaseCIFAR-10.
The third project is the binary and multi-class classification of images through convolutional neural networks (CNNs), using the Oxford-IIIT Pet Dataset database.
The binary classification task consists of distinguishing between images of dogs and cats, and the multi-class classification task consists of classifying images into one of 37 breeds of dogs and cats.
- For both tasks a CNN network was trained from scratch.
- A pre-trained Keras Network was chosen, and applied to the multi-class classification task.
- Data augmentation techniques were used.