Nonparametric Weight Initialization of Neural Networks via Integral Representation
This work addresses training efficiency for neural network practitioners, but it appears incremental as it builds on existing initialization techniques.
The authors tackled the problem of slow convergence in neural network training by proposing a nonparametric weight initialization method derived from integral representation, which led to faster backpropagation convergence and, in some cases, sufficient accuracy without further training.
A new initialization method for hidden parameters in a neural network is proposed. Derived from the integral representation of the neural network, a nonparametric probability distribution of hidden parameters is introduced. In this proposal, hidden parameters are initialized by samples drawn from this distribution, and output parameters are fitted by ordinary linear regression. Numerical experiments show that backpropagation with proposed initialization converges faster than uniformly random initialization. Also it is shown that the proposed method achieves enough accuracy by itself without backpropagation in some cases.