高性能计算和人工智能的实用量子加速.pdf

1、OCP EMEA Summit 29 April|Dublin,IrelandJames FletcherHead of Solutions Architecture,ORCA CPractical Quantum Acceleration for HPC and AIQuantum Computing with PhotonicsQuantum algorithms are built on quantum circuits:preparation,manipulation,measurementUniversal circuit with qubits and gatesNon-unive

2、rsal circuit with photons,qumodes and programmable interferenceQubits and gates is one way to do this,but not the only one!Sampling as a Computational PrimitiveBoson samplingRandom circuit samplingSending 40+photons into a circuit and measuring where they exit produces a classically hard distributio

3、nHighly non-uniform distribution:applications in ML and optimisationEntangling 50+qubits using a complex circuit and measuring their resulting state produces a classical hard distributionDistribution often close to uniform:mostly used for proofs of feasibilityQuantum processors can already generate

4、some distributions faster than classicalGoogle 2024:“10 septillion years on one of todays fastest supercomputers”When are Sampling Tasks Useful?OptimisationMachine learningEvaluating averagesHeuristic optimisation algorithms such as simulated annealing work by sampling from the space of solutions in

5、 a smart wayGenerative machine learning algorithms are sampling algorithms,trained to sample from a desirable data distributionEvaluating complex cost functions by averaging over many samples:applications in Monte Carlo sampling and classificationAbility to sample from new types of distributions can

6、 be helpful in:Example:Hybrid Quantum-Classical GANDiscriminatorGenerateddata XValidInvalidEmpirical data distributionEthernet,GigE NetworkClassical HPCPhotonic quantum processorReplace the classical“latent space”in a Generative Adversarial Network with a quantum latent spaceThe required quantum pro