Collaborative project with my friend Alexandre Hannebelle from master MVA to classify mnist using Kernel Ridge Regression. The code is inspired from these slides in the context of the course Kernel Methods in Machine Learning(Julien Mairal (INRIA)- Jean-Philippe Vert (ENS Paris & Mines ParisTech)), of the MVA master of ENS Cachan.
You can clone it from GitHub as follows:
$ git clone https://github.com/ThibaultGROUEIX/KernelMethods_mva.git
The data is provided with the repo.
Just run start.m and you're set !
Validation set | Test set |
---|---|
1000 images | 10 000 images |
100% | 97.78% |
- HoGs feature vector concatenating cell blocks of 4x4, 7x7 and 14x14. Gradients are put in 12 bins each time.
- Jittering
- Gaussian kernel (the best among the various kernel we tested)