Nowadays ‘build and run’ a predictive model is a quite easy task, thanks to frameworks that simplify things and set good defaults for you (i.e. Keras).
But how do you effectively train a model, in order to gain better performance or to get your results faster? Do you feel frustrated every time you need to set and then tune the network’s hyperparameters, too? Don’t worry!
In this talk I will share some practical tips&tricks (such as Model Ensembling and learning rate schedulers) and relative examples, derived from my personal experience or from literature, with the aim to improve neural networks capabilities and to get the convergence faster.
This talk is aimed at data scientists or everyone passionate about this topic who wants to learn more.
Gently inspired by Practical Deep Learning for Coders Part 1 v2 of fast.ai.
in __on sabato 21 aprile at 12:45 **See schedule**