News for hackers. Stuff that matters.
Re:Programming My Child by AstralStorm on Wednesday August 29, 2018 @04:00AM (#17866540)
The property you're looking for is called robustness. An algorithm that works with predictable results from many initial starting conditions without failure modes is called robust.

Biological systems, unlike computers, are usually extremely robust. It is not just one feedback, it is a highly complex set of multidimensional interacting feedback loops all around. Randomness and chaos are also crucial in some of these systems. And most of them actually are coded and not trained - but coded to be robust and adaptive.

The whole staging of development is likely either a side effect of this property or an additional cause of it.

You cannot truly reset any nontrivial (Esp. Nonlinear) unknown system with feedback where you don't have the direct access to states. This has nothing to do with training and everything to do with control theoretic property of observability.

Re:Programming My Child by j7ake on Wednesday August 29, 2018 @07:22AM (#17867344)
The property you want is beyond robust. You actually want the algorithm to constantly go beyond its comfortable prediction range and "fail" repeatedly. That's how you learn.
Re:Programming My Child by ericdykstra on Wednesday August 29, 2018 @07:57AM (#17867513)
Indeed, not just not fragile, not just robust, but you might even say antifragile: gaining from disorder.
Re:Programming My Child by AstralStorm on Wednesday August 29, 2018 @12:50PM (#17869710)
Sometimes, but that property is hard to actually ensure. (and partly why so many people break down too - their cognition is lacking in it)

Most of the bodily biological systems are not antifragile or are to a very small degree.