Timezone: »

 
Implicit Bias of Gradient Descent on Reparametrized Models: On Equivalence toMirror Descent
Zhiyuan Li · Tianhao Wang · Jason Lee · Sanjeev Arora

Sat Jul 23 08:40 AM -- 08:50 AM (PDT) @
in Continuous Time Perspectives in Machine Learning »

As part of the effort to understand implicit bias of gradient descent in overparametrized models, several results have shown how the training trajectory on the overparametrized model can be understood as mirror descent on a different objective. The main result here is a characterization of this phenomenon under a notion termed commuting parametrization, which encompasses all the previous results in this setting. It is shown that gradient flow with any commuting parametrization is equivalent to continuous mirror descent with a related mirror map. Conversely, continuous mirror descent with any mirror map can be viewed as gradient flow with a related commuting parametrization. The latter result relies upon Nash's embedding theorem.

Author Information

Zhiyuan Li (Princeton University)
Tianhao Wang (Yale University)
Jason Lee (Princeton)
Sanjeev Arora (Princeton University)

More from the Same Authors