Probabilistic bounds on neuron death in deep rectifier networks
This work addresses the problem of neuron death for researchers and practitioners in deep learning, providing theoretical insights and practical methods, though it is incremental in building on existing understanding of initialization issues.
The authors derived probabilistic bounds on the probability that a ReLU network is initialized to a trainable point, showing that depth can increase indefinitely with sufficient width, and proposed a sign flipping scheme guaranteeing at least a 2^{-k} ratio of living data points in a k-layer network.
Neuron death is a complex phenomenon with implications for model trainability: the deeper the network, the lower the probability of finding a valid initialization. In this work, we derive both upper and lower bounds on the probability that a ReLU network is initialized to a trainable point, as a function of model hyperparameters. We show that it is possible to increase the depth of a network indefinitely, so long as the width increases as well. Furthermore, our bounds are asymptotically tight under reasonable assumptions: first, the upper bound coincides with the true probability for a single-layer network with the largest possible input set. Second, the true probability converges to our lower bound as the input set shrinks to a single point, or as the network complexity grows under an assumption about the output variance. We confirm these results by numerical simulation, showing rapid convergence to the lower bound with increasing network depth. Then, motivated by the theory, we propose a practical sign flipping scheme which guarantees that the ratio of living data points in a $k$-layer network is at least $2^{-k}$. Finally, we show how these issues are mitigated by network design features currently seen in practice, such as batch normalization, residual connections, dense networks and skip connections. This suggests that neuron death may provide insight into the efficacy of various model architectures.