Project 2: Vaibhav Arcot

[Black-Phoenix]

• Repo Link
• Stream compation
  • Got stream compaction, scan working
  • Implemented RADIX sort
  • Implemented shared memory scan (and compared times)
  • Collected 34k (* 9 algorithms) datapoints over all permutations
• Character recognition
  • Achieved 100% accuracy on the test data.
  • Implemented SGD and adaptive learning rate algorithms.
  • Vectorized code entirely (matrix multiplications)
  • Trained the network on randomly rotated images of the fonts (100% accuracy)
  • Attempted 74k char dataset

[Black-Phoenix]

I am not sure if this code will be run to grade, but if it is, the number of output values (classes) will need to be changed from 26 to 52 (or however many the dataset requires) (It is a #define in main.cu). When will the grading be completed, because I would like to make the code more robust (add a few asserts and stuff) without it being considered during the grading process?