Media AI In Actuarial Science - the State of the Art

AI In Actuarial Science - the State of the Art

uploaded April 5, 2021 Views: 38 Comments: 1 Favorite: 2 CPD
Speakers: 
Tags:
Content groups:  content2021

1 Comments

Elsa
Elsa

1009 Days ago

I am referring to slide 45 about stability of the results of neural networks: there are papers discussing regularization techniques in deep networks (e.g. Sathya et al 2019) which promise increased stability, accuracy and training speed. It comes to me at no surprise that "problems are worse on small datasets" and it really seems to me a problem caused by too many parameters akin to having multiple solutions in a linear regression model with too many variables for a given number of observations..