Reproduce “Quantum Deep Learning Still Needs a Quantum Leap” on OpenResearch

About this paper

Hans Gundlach, Hrvoje Kukina, Jayson Lynch, Neil Thompson

This paper quantitatively assesses the practical impact of quantum computing on deep learning over the next one to two decades. It concludes that current quantum hardware and algorithms, even with rapid advancements, are unlikely to provide meaningful speedups due to immense constant factor overheads, reliance on undeveloped QRAM, and restrictive algorithmic conditions.

Key insights

• Quantum algorithms with only polynomial speedups are largely overwhelmed by the significant constant factor slowdown (estimated at ~10^13) of quantum hardware operations compared to classical ones.
• Many promising quantum algorithms for data-intensive deep learning tasks critically depend on Quantum Random Access Memory (QRAM), which remains an underdeveloped and challenging technology.
• Algorithms offering exponential speedups often require highly specialized problem conditions (e.g., sparse matrices, specific input vector preparation, returning only a quantum state) that are generally not met by typical deep learning applications.

Summary by alphaXiv.