Search papers, labs, and topics across Lattice.
Centre CMP, Mines Saint-Etienne, CEA-Leti, Equipe commune SAS
2
0
4
Achieving a 50% reduction in memory usage for on-device training without significant performance loss could revolutionize deep learning on resource-constrained hardware.
A mere 10 induced faults per malicious client can achieve a staggering 94% success rate in backdooring federated learning models.
