On the stability of learning with sparse and noisy gradients
The work shows that when gradient information is noisy and intermittently available, learning algorithms may become unstable or ineffective.
The work shows that when gradient information is noisy and intermittently available, learning algorithms may become unstable or ineffective.