Timeline for What makes parallel/distributed probabilistic inference difficult to implement?
Current License: CC BY-SA 3.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 29, 2016 at 18:45 | comment | added | jaradniemi | In my first draft, I had serial and changed to iterative since I was using the term iteration later. Updated the answer to use serial. | |
| Mar 29, 2016 at 18:44 | history | edited | jaradniemi | CC BY-SA 3.0 |
iterative -> serial based on comment
|
| Mar 29, 2016 at 16:19 | vote | accept | Bar | ||
| Mar 29, 2016 at 16:18 | comment | added | Bar | > "Parallelism in MCMC is hard because MCMC is inherently an iterative algorithm." I guess this hits the nail on the head then, although "serial" instead of "iterative" might be a better word here, since in my mind at least, there many iterative algorithms that are straight-forward to parallelize (like SGD), given certain assumptions (like convexity of the function). Are there relaxations like convexity that allow us to perform MCMC in parallel? | |
| Mar 29, 2016 at 14:46 | history | answered | jaradniemi | CC BY-SA 3.0 |