In this project I have implemented and tested neural network models for solving unsupervised problems. This is based on images of handwritten digits (MNIST). The basic tasks is to create, test and analyze a convolutional autoencoder. I have explored the use of advanced optimizers and regularization methods and hyperparameters have been tuned using appropriate search procedures. Final accuracy has be evaluated using a cross-validation setup.
- Implement and test (convolutional) autoencoder, reporting the trend of reconstruction loss
- Explore advanced optimizers and regularization methods
- Optimize hyperparameters using grid/random search and cross-validation
- Implement and test denoising (convolutional) autoencoder
- Fine-tune the (convolutional) autoencoder using a supervised classification task
- Explore the latent space structure (PCA) and generate new samples from latent codes
- Implement variational (convolutional) autoencoder
A detailed report is present in the nn_homework_02.pdf file