11 papers
StableGrad: Backward Scale Control without Batch Normalization
Jose I. Mestre, Alberto Fernández-Hernández, Cristian Pérez-Corral +2
Training very deep neural networks requires controlling the propagation of magnitudes across depth. Without such control, activations and gradients may vanish, explode, or enter un…
FedOUI: OUI-Guided Client Weighting for Federated Aggregation
Alberto Fernández-Hernández, Jose I. Mestre, Cristian Pérez-Corral +3
Federated learning usually aggregates client updates using dataset size or gradient-level criteria, while overlooking internal signals about how each client model is organizing its…
OUI as a Structural Observable: Towards an Activation-Centric View of Neural Network Training
Alberto Fernández-Hernández, Jose I. Mestre, Cristian Pérez-Corral +3
Activation functions are what make deep networks expressive: without them, the model collapses to a linear map. Yet we still evaluate training mostly from the outside, through loss…
Refresh-Scaling the Memory of Balanced Adam
Alberto Fernández-Hernández, Cristian Pérez-Corral, Jose I. Mestre +2
Recent evidence suggests that Adam performs robustly when its momentum parameters are tied, , reducing the optimizer to a single remaining parameter. However, how this p…
OUIDecay: Adaptive Layer-wise Weight Decay for CNNs Using Online Activation Patterns
Alberto Fernández-Hernández, Jose I. Mestre, Cristian Pérez-Corral +3
Weight decay remains one of the most widely used regularization mechanisms for training convolutional neural networks, yet it is still commonly applied as a fixed coefficient share…
Why Adam Works Better with : The Missing Gradient Scale Invariance Principle
Alberto Fernández-Hernández, Cristian Pérez-Corral, Jose I. Mestre +2
Adam has been at the core of large-scale training for almost a decade, yet a simple empirical fact remains unaccounted for: both validation scores and the qualitative behaviour of…