The project was about classifying MNIST data of low dimensions. After de-noising the reconstructed images using auto-encoders, I classified them using convolutional network and evaluated the performance. The images were reconstructed using PCA.
1. Prepare the MNIST dataset
2. Apply dimensionality reduction on the dataset using PCA ( variance = 0.90 )
3. Reconstruct the data from PCA ( the reconstructed data is noisy images )
4. Denoise the images using autoencoders
5. Train and test the denoised images on CNN model and evaluate the results.
6. Now, again apply dimensionality reduction using PCA with variance=0.50 and evaluate the resuls after training with CNN model.
Training Accuracy on PCA (variance = 0.90) = 98%
Training Accuracy on PCA (variance = 0.50) = 95%
Harsh Patel