« first day (946 days earlier)      last day (2177 days later) » 

forest
forest
12:53
Say, has Mentifex ever visited this stack?
I think it'd be really funny to see that.
(He's still active apparently)
 
6 hours later…
nbro
nbro
18:25
The following two questions should be moved to SO: ai.stackexchange.com/q/11065/2444 and ai.stackexchange.com/q/11068/2444.
Manuel Rodriguez
Manuel Rodriguez
18:54
@forest The question is the other way around: Is SE.AI familiar with the homepage of Mentifex? Have the users experience with the Forth programming language? For me the answer is no.
DuttaA
DuttaA
19:04
@NeilSlater hey take a look at this policy iteration I have done for a gridworld with terminal state at 0,0
Stochastic policy, Policy Evaluation step:
[[ 0. -5. -4.]
[-5. -4. -8.]
[-4. -8. -6.]]
Greedy step or policy Improvement Step:
[[ 0. -1. -6.]
[-1. -6. -5.]
[-6. -5. -9.]]
As you can see the value function will back and forth between these 2 states. What is wrong in my understanding?
this is after 10 iterations, the states are repeating
If we take the case of position 0,1 it has 6,6,0 values with -1,-1,-1 step cost, hence the policy evaluation step will give 0.333(-12-3) = -5
Whereas in the greedy step the best thing 0,1 can do is to go to 0,0 hence it will give value -6
and hence the states will oscillate. What am I missing here?
 
2 hours later…
DukeZhou
DukeZhou
20:50
"DeepMind is asking how Google helped turn the internet into an echo chamber" technologyreview.com/s/613083/…
nbro
nbro
@DukeZhou Yes, this issue isn't new. It's been around for quite a long time at this point
DukeZhou
DukeZhou
21:25
@nbro Well sure. And the alarm bells about cybersecurity and data privacy have been ringing for a couple of decades as well, without much action. (Here I just like the irony that DeepMind, a Google company, is doing the research;)

« first day (946 days earlier)      last day (2177 days later) »