A Multilayer Perceptron (MLP) Neural Network is trained using Pytorch LIghtning library. MNIST and Fashion MNIST datasets are used to trian the neural network.
Different Experiments are performed and results are observed. Type of experiments and validation accuracy of network is given below:
Version 0: Batch_SIze: 8 Hidden_Layers: 1 Hidden Neurons: 1024 Optimizer: SGD Sigmoid Val_Acc: 0.9527
Version 1: Batch_SIze: 8 Hidden_Layers: 1 Hidden Neurons: 512 Optimizer: SGD Sigmoid Val_Acc: 0.9552
Version 2: Batch_SIze: 16 Hidden_Layers: 1 Hidden Neurons: 512 Optimizer: SGD Sigmoid Val_Acc: 0.9377 Increasing
Version 3: Batch_SIze: 32 Hidden_Layers: 1 Hidden Neurons: 512 Optimizer: SGD Sigmoid Val_Acc: 0.9298 Increasing
Version 4: Batch_SIze: 32 Hidden_Layers: 1 Hidden Neurons: 512 Optimizer: SGD Relu Val_Acc: 0.9739
Version 5: Batch_SIze: 32 Hidden_Layers: 1 Hidden Neurons: 512 Optimizer: Adam Relu Val_Acc: 0.9763
Version 6: Batch_SIze: 32 Hidden_Layers: 1 Hidden Neurons: 512 Optimizer: Adam Relu Val_Acc: 0.9835 With Augmentation