To do
- Same as yesterday
- Go over virtual worlds multi object tracking paper with the reading group
Done
- Still working on Part 1. I managed to get the error down to 36.2%! This is without using a deep net.
- Started adding more layers to my network. It increases error up to ~39% due to overfitting
- We read the virtual worlds multi object tracking paper
- We went over kernels in-depth
- I finally get the intuition behind kernels but I need to check out the math for representing the minimization problem in terms of training data instead of features