The photonic Ising machine, a promising non-von Neumann computational paradigm, offers a feasible way to address combinatorial optimization problems. We develop a digital noise injection method for spatial photonic Is...The photonic Ising machine, a promising non-von Neumann computational paradigm, offers a feasible way to address combinatorial optimization problems. We develop a digital noise injection method for spatial photonic Ising machines based on smoothed analysis, where noise level acts as a parameter that quantifies the smoothness degree. Through experiments with 20736-node Max-Cut problems, we establish a stable performance within a smoothness degree of 0.04 to 0.07. Digital noise injection results in a 24% performance enhancement, showing a 73% improvement over heuristic Sahni–Gonzales (SG) algorithms. Furthermore, to address noise-induced instability concerns, we propose an optoelectronic co-optimization method for a more streamlined smoothing method with strong stability.展开更多
Combinatorial optimization problems and ground state problems of spin glasses are crucial in various fields of science and technology.However,they often belong to the computational class of NP-hard,presenting signific...Combinatorial optimization problems and ground state problems of spin glasses are crucial in various fields of science and technology.However,they often belong to the computational class of NP-hard,presenting significant computational challenges.Traditional algorithms inspired by statistical physics like simulated annealing have been widely adopted.Recently,advancements in Ising machines,such as quantum annealers and coherent Ising machines,offer new paradigms for solving these problems efficiently by embedding them into the analog evolution of nonlinear dynamical systems.However,existing dynamics-based algorithms often suffer from low convergence rates and local minima traps.In this work,we introduce the dual mean-field dynamics into Ising machines.The approach integrates the gradient force and the transverse force into the dynamics of Ising machines in solving combinatorial optimization problems,making it easier for the system to jump out of the local minimums and allowing the dynamics to explore wider in configuration space.We conduct extensive numerical experiments using the Sherrington–Kirkpatrick spin glass up to 10000 spins and the maximum cut problems with the standard G-set benchmarks.The numerical results demonstrate that our dual mean-field dynamics approach enhances the performance of base Ising machines,providing a more effective solution for large-scale combinatorial optimization problems.展开更多
Spatial photonic Ising machines,as emerging artificial intelligence hardware solutions by leveraging unique physical phenomena,have shown promising results in solving large-scale combinatorial problems.However,spatial...Spatial photonic Ising machines,as emerging artificial intelligence hardware solutions by leveraging unique physical phenomena,have shown promising results in solving large-scale combinatorial problems.However,spatial light modulator enabled Ising machines still remain bulky,are very power demanding,and have poor stability.In this study,we propose an integrated XY Ising sampler based on a highly uniform multimode interferometer and a phase shifter array,enabling the minimization of both discrete and continuous spin Hamiltonians.We elucidate the performance of this computing platform in achieving fully programmable spin couplings and external magnetic fields.Additionally,we successfully demonstrate the weighted full-rank Ising model with a linear dependence of 0.82 and weighted MaxCut problem solving with the proposed sampler.Our results illustrate that the developed structure has significant potential for larger-scale,reduced power consumption and increased operational speed,positioning it as a versatile platform for commercially viable high-performance samplers of combinatorial optimization problems.展开更多
In recent years,the explosive development of artificial intelligence implementing by artificial neural networks(ANNs)creates inconceivable demands for computing hardware.However,conventional computing hardware based o...In recent years,the explosive development of artificial intelligence implementing by artificial neural networks(ANNs)creates inconceivable demands for computing hardware.However,conventional computing hardware based on electronic transistor and von Neumann architecture cannot satisfy such an inconceivable demand due to the unsustainability of Moore’s Law and the failure of Dennard’s scaling rules.Fortunately,analog optical computing offers an alternative way to release unprecedented computational capability to accelerate varies computing drained tasks.In this article,the challenges of the modern computing technologies and potential solutions are briefly explained in Chapter 1.In Chapter 2,the latest research progresses of analog optical computing are separated into three directions:vector/matrix manipulation,reservoir computing and photonic Ising machine.Each direction has been explicitly summarized and discussed.The last chapter explains the prospects and the new challenges of analog optical computing.展开更多
Quantum computing is an emerging technology that is expected to realize an exponential increase in computing power. Recently,its theoretical foundation and application scenarios have been extensively researched and ex...Quantum computing is an emerging technology that is expected to realize an exponential increase in computing power. Recently,its theoretical foundation and application scenarios have been extensively researched and explored. In this work, we propose efficient quantum algorithms suitable for solving computing power scheduling problems in the cloud-rendering domain, which can be viewed mathematically as a generalized form of a typical NP-complete problem, i.e., a multiway number partitioning problem.In our algorithm, the matching pattern between tasks and computing resources with the shortest completion time or optimal load balancing is encoded into the ground state of the Hamiltonian;it is then solved using the optical coherent Ising machine, a practical quantum computing device with at least 100 qubits. The experimental results show that the proposed quantum scheme can achieve significant acceleration and save 97% of the time required to solve combinatorial optimization problems compared with classical algorithms. This demonstrates the computational advantages of optical quantum devices in solving combinatorial optimization problems. Our algorithmic and experimental work will advance the utilization of quantum computers to solve specific NP problems and will broaden the range of possible applications.展开更多
基金supported by the National Natural Science Foundation of China(Nos.62235011 and 62175146).
文摘The photonic Ising machine, a promising non-von Neumann computational paradigm, offers a feasible way to address combinatorial optimization problems. We develop a digital noise injection method for spatial photonic Ising machines based on smoothed analysis, where noise level acts as a parameter that quantifies the smoothness degree. Through experiments with 20736-node Max-Cut problems, we establish a stable performance within a smoothness degree of 0.04 to 0.07. Digital noise injection results in a 24% performance enhancement, showing a 73% improvement over heuristic Sahni–Gonzales (SG) algorithms. Furthermore, to address noise-induced instability concerns, we propose an optoelectronic co-optimization method for a more streamlined smoothing method with strong stability.
基金supported by Projects 12325501,12047503,12247104,and 12322501 of the National Natural Science Foundation of ChinaProject ZDRW-XX-2022-302 of the Chinese Academy of Sciencespartially supported by the Innovation Program for Quantum Science and Technology project 2021ZD0301900。
文摘Combinatorial optimization problems and ground state problems of spin glasses are crucial in various fields of science and technology.However,they often belong to the computational class of NP-hard,presenting significant computational challenges.Traditional algorithms inspired by statistical physics like simulated annealing have been widely adopted.Recently,advancements in Ising machines,such as quantum annealers and coherent Ising machines,offer new paradigms for solving these problems efficiently by embedding them into the analog evolution of nonlinear dynamical systems.However,existing dynamics-based algorithms often suffer from low convergence rates and local minima traps.In this work,we introduce the dual mean-field dynamics into Ising machines.The approach integrates the gradient force and the transverse force into the dynamics of Ising machines in solving combinatorial optimization problems,making it easier for the system to jump out of the local minimums and allowing the dynamics to explore wider in configuration space.We conduct extensive numerical experiments using the Sherrington–Kirkpatrick spin glass up to 10000 spins and the maximum cut problems with the standard G-set benchmarks.The numerical results demonstrate that our dual mean-field dynamics approach enhances the performance of base Ising machines,providing a more effective solution for large-scale combinatorial optimization problems.
基金National Natural Science Foundation of China(62175146,62235011)。
文摘Spatial photonic Ising machines,as emerging artificial intelligence hardware solutions by leveraging unique physical phenomena,have shown promising results in solving large-scale combinatorial problems.However,spatial light modulator enabled Ising machines still remain bulky,are very power demanding,and have poor stability.In this study,we propose an integrated XY Ising sampler based on a highly uniform multimode interferometer and a phase shifter array,enabling the minimization of both discrete and continuous spin Hamiltonians.We elucidate the performance of this computing platform in achieving fully programmable spin couplings and external magnetic fields.Additionally,we successfully demonstrate the weighted full-rank Ising model with a linear dependence of 0.82 and weighted MaxCut problem solving with the proposed sampler.Our results illustrate that the developed structure has significant potential for larger-scale,reduced power consumption and increased operational speed,positioning it as a versatile platform for commercially viable high-performance samplers of combinatorial optimization problems.
文摘In recent years,the explosive development of artificial intelligence implementing by artificial neural networks(ANNs)creates inconceivable demands for computing hardware.However,conventional computing hardware based on electronic transistor and von Neumann architecture cannot satisfy such an inconceivable demand due to the unsustainability of Moore’s Law and the failure of Dennard’s scaling rules.Fortunately,analog optical computing offers an alternative way to release unprecedented computational capability to accelerate varies computing drained tasks.In this article,the challenges of the modern computing technologies and potential solutions are briefly explained in Chapter 1.In Chapter 2,the latest research progresses of analog optical computing are separated into three directions:vector/matrix manipulation,reservoir computing and photonic Ising machine.Each direction has been explicitly summarized and discussed.The last chapter explains the prospects and the new challenges of analog optical computing.
基金supported by the National Key R&D Plan (Grant No. 2021YFB2801800)。
文摘Quantum computing is an emerging technology that is expected to realize an exponential increase in computing power. Recently,its theoretical foundation and application scenarios have been extensively researched and explored. In this work, we propose efficient quantum algorithms suitable for solving computing power scheduling problems in the cloud-rendering domain, which can be viewed mathematically as a generalized form of a typical NP-complete problem, i.e., a multiway number partitioning problem.In our algorithm, the matching pattern between tasks and computing resources with the shortest completion time or optimal load balancing is encoded into the ground state of the Hamiltonian;it is then solved using the optical coherent Ising machine, a practical quantum computing device with at least 100 qubits. The experimental results show that the proposed quantum scheme can achieve significant acceleration and save 97% of the time required to solve combinatorial optimization problems compared with classical algorithms. This demonstrates the computational advantages of optical quantum devices in solving combinatorial optimization problems. Our algorithmic and experimental work will advance the utilization of quantum computers to solve specific NP problems and will broaden the range of possible applications.
