SOCR ≫ DSPA ≫ Topics ≫

1 Deep learning Classification

  • Download the Alzheimer’s data from the SOCR Archive.
  • Properly preprocess the data and remove outliers.
  • Build a multi-layer perceptron as a classifier and select proper parameters.
  • Classify AD and NC and report the detailed classification accuracty metrics using cross table, accuracy, sensitivity, specificity, LOR, AUC.
  • Generate some data/results visualizations, at least include histograms and model graph structures. see Chater 22.
  • Try to construct a deeper and more elaborate network model and report the prediction results.
  • Compare your results with alternative data-driven methods (e.g., KNN).

2 Deep learning Regression

  • Download the Allometric relationship data from the SOCR data archive.
  • Preprocess the data and set density as the response variable.
  • Generate a MXNet feedforward neural net model and properly specify its parameters.
  • Train the model and use it to predict the response. Report RMSE on the test data, evaluate the results and justify your evaluation.
  • Output the model’s graph structure.

3 Image classification

Apply deep learning neural network models to classify the following images using the pre-trained model we showed in Chapter 22:

SOCR Resource Visitor number Dinov Email