US 11,922,337 B2
Accelerator for computing combinatorial cost function
Matthias Troyer, Clyde Hill, WA (US); Helmut Gottfried Katzgraber, Kirkland, WA (US); and Christopher Anand Pattison, College Station, TX (US)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Jan. 20, 2023, as Appl. No. 18/157,339.
Application 18/157,339 is a continuation of application No. 16/272,851, filed on Feb. 11, 2019, granted, now 11,562,273.
Prior Publication US 2023/0153665 A1, May 18, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 7/01 (2023.01); G06F 7/58 (2006.01); G06F 9/38 (2018.01); G06F 17/11 (2006.01)
CPC G06N 7/01 (2023.01) [G06F 7/582 (2013.01); G06F 9/3877 (2013.01); G06F 17/11 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A computing device comprising:
memory storing instructions to compute a combinatorial cost function of a plurality of variables;
an accelerator device; and
a processor configured to:
generate a plurality of data packs, wherein each data pack indicates an update to a variable of the one or more variables; and
transmit the plurality of data packs to the accelerator device;
wherein the accelerator device is configured to:
for each data pack:
retrieve a variable value of the variable indicated by the data pack;
generate a pseudorandom number;
generate an updated variable value of the variable as indicated by the data pack;
generate an updated cost function value of the combinatorial cost function based on the updated variable value;
determine a transition probability based at least in part on the updated cost function value; and
store the updated variable value and the updated cost function value for the variable indicated in the data pack in response to determining that the transition probability exceeds the pseudorandom number; and
output a final updated cost function value of the combinatorial cost function to the processor.