Know What's Happening in Technology

Los Alamos Researchers Show Some Quantum Learning Models Are Classically Simulable

Los Alamos National Laboratory researchers published a perspective in Nature Communications showing that quantum machine‑learning models designed to avoid barren plateaus are often classically simulable. By restricting variational quantum circuits to small subspaces, they demonstrated end‑to‑end classical surrogates that match or exceed quantum convolutional neural networks on benchmark tasks, even for simulations involving up to 1,024 qubits. The work suggests that the very tricks used to sidestep training flatness may erase any quantum advantage. A hybrid proposal—using quantum hardware to generate data for classical algorithms—offers a possible path forward.