qubits qoptools directions · • map problem to a higher-order binary optimization (hubo) on an...
TRANSCRIPT
Qubits
SteveReinhardtDirectorofSoftwareToolsD-WaveSystems
qOp ToolsDirections
©2016D-WaveSystemsInc.AllRightsReserved Qubits
TheqOp goalsaretoestablishkeyabstractionsthatarevaluableforapplicationsandhigher-leveltoolsandeffectivelyexecutethemon
D-Wavesystems.
©2016D-WaveSystemsInc.AllRightsReserved Qubits
qOp Tools
• Initiallydevelopedasprototypestospuruserengagement• qbsolv:HybridpartitioningoptimizationtoolsolvesavirtualQUBO*• ToQ:Constraint-satisfactionlanguageandsolver• dw:Command-lineinterfacetoSAPI• qsage:Optimizationsolversolvesaninputobjectivefunction• QuantumApprentice(QuApp):MS-Excel-basedtrainingtoolenablesexperimentationwithsmallnumbersofqubits
• qbqual:ToolthatqualifiesaQUBOforeffectiveexecutiononaD-Wavesystem
*Quadraticunconstrainedbinaryoptimizationproblem
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Givens• Strategies• FocusareasandRoadmap
Outline
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• TheD-Wavesystemwillevolveandsoondeliverquantumadvantage
• Differingtypesofuserswillhavedifferingtoolrequirements• Subject-matterexperts(SMEs)arealreadysolvingD-Wave-styleproblemstodaywithexistingtools
• D-Waveisnottheonlyorganizationdevelopingtools;needtobeagoodcitizen
• AlmostallproblemsjustifyingD-Waveaccesswillbebiggerthanhardware
Givens
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Bilingual:conversantwithhardwareandapplicationlayers– Deephardwareand domainknowledgerequired
– Potentiallynecessaryforfirstfewgainfulapps
– Scarce
• Monolingual:conversantwithapplicationlayers– Willingtoreworkapplicationtonewformulation
– Notwillingtolearn(much)quantumannealingphysics
– Muchmoreabundant
TypesofInnovatorUsers
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Bilingual– Easierisnice,butdon’tgetinthewayofany potentiallyvaluablecapability
• Monolingual– Ifatallpractical,letmekeepusingmydomain- ormethod-specificinterface
andhavethatmaptoD-Wave
– IfImustchangetoanewprogramminginterface,thatisabigdeal,sothenewinterfacemustlastandevolvealongtime
– Shieldmefromnearlyallthedetailsofmappingtoagivensystem
ToolRequirements
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Givens• Strategies• FocusareasandRoadmap
Agenda
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• EstablishstrongabstractionstofosterapplicationsandSME*-focusedlayeredtools– Enablesconcurrentdevelopmentbyhardware,tools,andappdevelopers
• Addvalueclosertothesystem• Deliverandrespondquickly– Distinguishbetweenprototypesandproducts
• Collaboratewithcapablepartnersviaopen-sourcewhenappropriate
• Whenquantumadvantageisreal,deliverittomultipledomainsquickly
Strategies
*Subject-matterexpert
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Mapproblemtoahigher-orderbinaryoptimization(HUBO)onanenergyscale
• MapHUBOtoaQUBO• OptimizetheQUBO• Partitiontofitonhardwareafterfollowingsteps• Mapfromvirtualtophysicalconnectivity• Scaletolimitednumericalrange• Controlforhardwareskew• Micro-adjustannealingforbestresults• Correctforquantumerrors
OneApproachtoUsingtheQMI*
*QuantumMachineInstruction
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Mapproblemtoahigher-orderbinaryoptimization(HUBO)onanenergyscale
• MapHUBOtoaQUBO
• OptimizetheQUBO• Partitiontofitonhardwareafterfollowingsteps• Mapfromvirtualtophysicalconnectivity• Scaletolimitednumericalrange• Controlforhardwareskew• Micro-adjustannealingforbestresults• Correctforquantumerrors
OneApproachtoUsingtheQMI
Method-dependent
Method-independent
VirtualQUBO
©2016D-WaveSystemsInc.AllRightsReserved Qubits
VirtualQUBOAbstraction
Abstraction
qbsolv
https://pixabay.com/en/superhero-human-hero-woman-female-152840/
… Modelingframeworks
Istheabstractionhighlyuseful?
VirtualQUBO
Isiteffectivelyimplementable?
QuantumMachineInstruction
TargetQuantumProcessors
QMISAPI(C,C++,MATLAB,Python)SolverAPI(C,C++,MATLAB,Python)
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• qbsolv partitionsavirtualQUBOforD-Waveexecution– AlgorithmbasedonWang,Lu,Glover,andHao [2012]– Numericalresultscompetitivewithbestotherresults;executiontime(solelyclassical)3Xfaster
– SolvesproblemsO(10X)biggerthanunderlyingD-Wavesystem– CollaboratingwithGlover,Lewis,etal.onbetteralgorithm– Anticipatingopen-sourceavailabilitythisfall
• Otherworkonhybridpartitioningoptimizationsolvers– 1QBit:”Buildinganiterativeheuristicsolverforaquantumannealer”,http://arxiv.org/pdf/1507.07605v2.pdf
– Spedalieri,Albash etal.:seeHPECandthisafternoon’spresentations
• Effectivelyimplementable?:Sofar,Yes
EffectivelyImplementable?
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• KarpshowedthatallNPproblemsconvertibletootherNPproblems– Note:AndrewLucas(“Ising formulationsofmanyNPproblems”)isworkingwithus
• Gloveretalreportgoodsuccessconvertinghigh-valueproblemstoBQPs(==QUBOs)“TheUnconstrainedBinaryQuadraticProgrammingProblem:ASurvey”,http://leeds-faculty.colorado.edu/glover/454%20-%20xQx%20survey%20article%20as%20published%202014.pdf
• ToQ emitsvQUBO forlargeproblems– EnablesToQ tosolveproblems>hardware
– Works,butvQUBO notidealforconstraint-satisfactionproblems
• Highlyuseful?:Promising,butinsufficientdata
HighlyUsefulforSMETools?
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Givens• Strategies• FocusareasandRoadmap
Agenda
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• Mapproblemtoahigher-orderbinaryoptimization(HUBO)onanenergyscale
• MapHUBOtoaQUBO
• OptimizetheQUBO• Partitiontofitonhardwareafterfollowingsteps• Mapfromvirtualtophysicalconnectivity• Scaletolimitednumericalrange• Controlforhardwareskew• Micro-adjustannealingforbestresults• Correctforquantumerrors
OneApproachtoUsingtheQMI
Method-dependent
Method-independent
VirtualQUBO
qbsolvPartitioningconstraint-sat
solver
executor
©2016D-WaveSystemsInc.AllRightsReserved Qubits
Product Stack
Solver&SamplingLibraries
QCSystem
StandardModellingFrameworks
CommonLanguages,IDEs,&APIs
Models
Sampling&OptimizationEngine
Applications
HybridSamplingMethods
SATFilterRBM
BoltzmannSamples
HybridOptimizationMethods
StructuredSVM
QUBOSolutions
©2016D-WaveSystemsInc.AllRightsReserved Qubits
Current qOp Components
Solver&SamplingLibraries
QCSystem
StandardModellingFrameworks
CommonLanguages,IDEs,&APIs
Models
Sampling&OptimizationEngine
ApplicationUsertraining
HybridOptimizationMethod(qbsolv)
qsage
QuApp,dw,qbqual
ToQ
©2016D-WaveSystemsInc.AllRightsReserved Qubits
• qOp toolsareprototypesfornow– Rapiddelivery,rapidfeedback,rapidevolution
– Notproductionquality
– Intime,somewilllikelybecomeproducts
• Packagesmadeavailable~quarterly• qOp 2.2madeavailableAug1– ToQ:experimentalversionadded
– qbsolv:performanceimprovements
• qOp 2.3targetedforlateOctober– qbsolv:open-source-ready,moreperformanceimprovements
Practicalities
©2016D-WaveSystemsInc.AllRightsReserved Qubits
qOp Roadmap – Spring 2017
Solver&SamplingLibraries
StandardModellingFrameworks
Sampling&OptimizationEngine
HybridOptimizationSolver(qbsolv)
qsage
ToQ
executor
ConstraintSatisfactione.g., Cryptol,MiniZinc
vQUBO
©2016D-WaveSystemsInc.AllRightsReserved Qubits
qOp Roadmap – Fall 2017
CoreLibraries
StandardModellingFrameworks
Sampling&OptimizationEngine
HybridOptimizationSolver(qbsolv)
qsage
ToQ
executor
HybridConstraint-SatisfactionSolver
ConstraintSatisfactione.g., Cryptol,MiniZinc
vQUBO <???>
©2016D-WaveSystemsInc.AllRightsReserved Qubits
TheqOp goalsaretoestablishkeyabstractionsthatarevaluableforapplicationsandhigher-leveltoolsandeffectivelyexecutethemon
D-Wavesystems.
©2016D-WaveSystemsInc.AllRightsReserved Qubits
Backup
©2016D-WaveSystemsInc.AllRightsReserved Qubits
Wherearewegoing?
Ourgoalistoenableourcustomerstobesuccessful.
Wewanttheproducttoenablemoreuserstobuildeffectivealgorithms,modelsandapplications.
Sampling&Optimization
Engine
Models
Applications
Sample/SolutionQuality
GraphQuality
Sample/SolutionRate