Just Pick a Sign: Optimizing Deep Multitask Models with Gradient Sign Dropout

by   Zhao Chen, et al.

The vast majority of deep models use multiple gradient signals, typically corresponding to a sum of multiple loss terms, to update a shared set of trainable weights. However, these multiple updates can impede optimal training by pulling the model in conflicting directions. We present Gradient Sign Dropout (GradDrop), a probabilistic masking procedure which samples gradients at an activation layer based on their level of consistency. GradDrop is implemented as a simple deep layer that can be used in any deep net and synergizes with other gradient balancing approaches. We show that GradDrop outperforms the state-of-the-art multiloss methods within traditional multitask and transfer learning settings, and we discuss how GradDrop reveals links between optimal multiloss training and gradient stochasticity.


page 1

page 2

page 3

page 4


Multitask Learning with Single Gradient Step Update for Task Balancing

Multitask learning is a methodology to boost generalization performance ...

GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks

Deep multitask networks, in which one neural network produces multiple p...

MTAdam: Automatic Balancing of Multiple Training Loss Terms

When training neural models, it is common to combine multiple loss terms...

Deep Counterfactual Networks with Propensity-Dropout

We propose a novel approach for inferring the individualized causal effe...

Master Thesis: Neural Sign Language Translation by Learning Tokenization

In this thesis, we propose a multitask learning based method to improve ...

Neural Sign Language Translation by Learning Tokenization

Sign Language Translation has attained considerable success recently, ra...

Multitask and Transfer Learning for Autotuning Exascale Applications

Multitask learning and transfer learning have proven to be useful in the...