The Deep Weight Prior arXiv:1810.06943
We are going to reproduce considered in the paper experiments and prove (or disprove) that for dwp we have that
- dwp improve the performance of Bayesian neural networks in case of limited data
- initialization of weights with samples from dwp accelerates training of conventional convolutional neural networks, such that training procedure becomes faster than, say, in uniform initialization case, or in Xavier initialization case
- Leonid Matyushin
- run Auxiliary CNN Training with the command
python get_kernels.py
- run Experiments with the command
runipy experiments.ipynb
. You can install runipy viapip install runipy