ML-experiments

A few basic neural networks and my experiments in deep learning

Convolutional Neural Networks

One Convolutional Neural Network (CNN) has been implemented and tried on the MNIST and CIFAR-10 Datasets with the results:-

• MNIST Dataset - 98.26% Accuracy and 98.24% F-Measure
• CIFAR-10 Dataset
  • 58.1% Accuracy and 48% F-Measure in 25 Epochs (Probably still underfitting at that point)
  • 79.3% Accuracy and 78.8% F-Measure in 100 Epochs

A deep CNN was implemented on the CIFAR-10 Dataset

• CIFAR-10 Deep CNN - 44.17% Accuracy and 26% F-Measure in 100 Epochs (Either overfitting, or a gradient overshoot)