<ul><li>Deep neural networks are not resilient to parameter corruptions: even a single-bitwise error in their parameters in memory can cause an accuracy drop of over 10%.</li><li>Hessian-aware training is proposed as an approach to improve resilience to bitwise corruptions in neural network parameters.</li><li>The approach promotes models with flatter loss surfaces and shows a reduction in the number of bits leading to a significant accuracy drop.</li><li>This method can work synergistically with existing hardware and system-level defenses.</li></ul>

Hessian-aware Training for Enhancing DNNs Resilience to Parameter Corruptions

Discover more