This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
security [2018/10/24 17:04]
security [2019/01/12 10:59] (current)
Line 72: Line 72:
 https://​ai.google/​education/​responsible-ai-practices?​twitter=@bigdata https://​ai.google/​education/​responsible-ai-practices?​twitter=@bigdata
 +https://​arxiv.org/​abs/​1812.00564v1 Split learning for health: Distributed deep learning without sharing raw patient data
 +https://​arxiv.org/​abs/​1806.01186 Measuring and avoiding side effects using relative reachability
 + We introduce a general definition of side effects, based on relative reachability of states compared to a default state, that avoids these undesirable incentives. Using a set of gridworld experiments illustrating relevant scenarios, we empirically compare relative reachability to penalties based on existing definitions and show that it is the only penalty among those tested that produces the desired behavior in all the scenarios.