Multi-layer Neural Network
- 1 min
Neural networks are considered to be the state of the art of technologies, and in the center lies the network’s ability to learn from its mistakes. What is more, the configurability of this epitome of human creation allows the technology to be versatile enough to be used in different settings. In this report, the team inves- tigates how back propagation, when paired with momentum, regularization, and various activation functions, helps neural network improve its accuracy in iden- tifying handwritten Japanese characters. With the best model we were able to get 88.18% test accuracy with following configurations [’layer specs’: [784, 512, 10], ’activation’: ’ReLU’, ’learning rate’: 0.0005, ’batch size’: 128, ’epochs’: 110, ’early stop’: True, ’early stop epoch’: 5, ’L2 penalty’: 0.0001, ’momentum’: True, ’momentum gamma’: 0.9].