Search for: All records

Our study evaluates the limitations and potentials of Quantum Random Access Memory (QRAM) within the principles of quantum physics and relativity. QRAM is crucial for advancing quantum algorithms in fields like linear algebra and machine learning, purported to efficiently manage large data sets with$${{{\mathcal{O}}}}(\log N)$$$O\left(\log N\right)$circuit depth. However, its scalability is questioned when considering the relativistic constraints on qubits interacting locally. Utilizing relativistic quantum field theory and Lieb–Robinson bounds, we delve into the causality-based limits of QRAM. Our investigation introduces a feasible QRAM model in hybrid quantum acoustic systems, capable of supporting a significant number of logical qubits across different dimensions-up to ~10⁷in 1D, ~10¹⁵to ~10²⁰in 2D, and ~10²⁴in 3D, within practical operation parameters. This analysis suggests that relativistic causality principles could universally influence quantum computing hardware, underscoring the need for innovative quantum memory solutions to navigate these foundational barriers, thereby enhancing future quantum computing endeavors in data science.

 

Large machine learning models are revolutionary technologies of artificial intelligence whose bottlenecks include huge computational expenses, power, and time used both in the pre-training and fine-tuning process. In this work, we show that fault-tolerant quantum computing could possibly provide provably efficient resolutions for generic (stochastic) gradient descent algorithms, scaling as$${{{{{{{\mathcal{O}}}}}}}}({T}^{2}\times {{{{{{{\rm{polylog}}}}}}}}(n))$$$O\left(T^2\times \mathrm{polylog}\left(n\right)\right)$, wherenis the size of the models andTis the number of iterations in the training, as long as the models are both sufficiently dissipative and sparse, with small learning rates. Based on earlier efficient quantum algorithms for dissipative differential equations, we find and prove that similar algorithms work for (stochastic) gradient descent, the primary algorithm for machine learning. In practice, we benchmark instances of large machine learning models from 7 million to 103 million parameters. We find that, in the context of sparse training, a quantum enhancement is possible at the early stage of learning after model pruning, motivating a sparse parameter download and re-upload scheme. Our work shows solidly that fault-tolerant quantum algorithms could potentially contribute to most state-of-the-art, large-scale machine-learning problems.

 

Hybridizing different degrees of freedom or physical platforms potentially offers various advantages in building scalable quantum architectures. Here, we introduce a fault-tolerant hybrid quantum computation by building on the advantages of both discrete-variable (DV) and continuous-variable (CV) systems. In particular, we define a CV-DV hybrid qubit with a bosonic cat code and a single photon, which is implementable in current photonic platforms. Due to the cat code encoded in the CV part, the predominant loss errors are readily correctable without multiqubit encoding, while the logical basis is inherently orthogonal due to the DV part. We design fault-tolerant architectures by concatenating hybrid qubits and an outer DV quantum error-correction code such as a topological code, exploring their potential merit in developing scalable quantum computation. We demonstrate by numerical simulations that our scheme is at least an order of magnitude more resource efficient compared to all previous proposals in photonic platforms, allowing us to achieve a record-high loss threshold among existing CV and hybrid approaches. We discuss the realization of our approach not only in all-photonic platforms but also in other hybrid platforms including superconducting and trapped-ion systems, which allows us to find various efficient routes toward fault-tolerant quantum computing.