The Project 2 breakdown based on the subprojects is as follows: 60% Stream Compaction, 40% Character Recognition
Since Stream compaction was for demonstrating a strong understanding of threading as a whole; whereas Character Recognition is a demonstration of some applications of it (as will be a lot of future 565 projects).
Required:
- CODE (as always)
- If you submit the baseline for the MLP project - submit a txt file of the weights you used and how close your final result was in terms of error / accuracy for each individual character
- If you submit an extra credit or your own version of the MLP project - clearly state what changes you did, how your network / convolution is organized, and what weights / setup you used (make your readme as clear as possible)
for EC, there will be 20pts max for this assignment.
for character recognition - main method contains the stream compaction test as an example, note that this can be deleted before submission (comment it out for hw release)
look at
to provide more description for features and implementation (other than just the xor excel sheet)