Investigation enlargement might help to some extent, however it is impractical to expect everything
Finally, info is queen. In case the education data doesn’t satisfy the take to analysis, you can show all you want nonetheless rating scrap performance. Possibly assemble sufficient degree data to fund every shot cases or, in the event that’s not possible from the start, retrain having the latest data regularly.
Concurrently, the fresh optimizer really does in fact appear to have a variety of energy, even after states actually stating the contrary, and you may spends it having a great nesterov-instance action (line dos off step three on the internal circle).