search

accosmin / nano

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

Layers should use floating point tensors in their interface #12

Closed accosmin closed 9 years ago

accosmin commented 9 years ago

This way the layers can be 2 times faster, because of float-based vectorization of various linear operations.

accosmin commented 9 years ago

Moved to 0.3 as it may require significant adjustments in the accuracy tests & the optimization module.

accosmin commented 9 years ago

Using double in the end as using floats impacts negatively the numerical optimizers.