Supervised learning
- load mnist dataset
- try Random Forest and Gradient Boosted Trees to classify mnist data set
- perform crossvalidation and grid search over the parameters of RF and GBT
- choose neural network library and fit mnist dataset
- build and train network similar to LeNet, NeoCognitron or some other specially design for digit recognition (advanced)
Unsupervised learning
- try several clustering algorithms on mnist dataset (how to estimate quality of the clustering considering we know labels)
- perform PCA + any classifier algorithm
- perform PCA + Decision Tree, compare results to Decision Tree alone for different maximal tree heights.
- perform Isomap (or other manifold learning) on mnist data set, plot the result mapping (advanced)
- Go to http://www.scala-lang.org/ and install scala. Make sure scala shell (
scala command) works fine.
- Get familiar with scala programming language (advanced)