search

accosmin / nano

C++ library [machine learning & numerical optimization] - superseeded by libnano
MIT License
1 stars 0 forks source link

Investigate ADADELTA #78

Closed accosmin closed 8 years ago

accosmin commented 8 years ago

The weighting scheme does not match the one introduced in the original paper.

accosmin commented 8 years ago

May tune the momentum factor for the "averaging" methods (adadelta, adagrad, sia, sga).

accosmin commented 8 years ago

Could try to tune momentum like {0.1, 0.2, 0.5, 0.9, 0.95}