BCIwiki (bciwiki.com) — A preprint study integrates parameterized quantum circuits (PQCs) as residual sidecar modules within a ResNet-50 backbone for 31-class neural population decoding, classifying imagined handwriting from multi-neuron spike rasters. Under strictly controlled conditions, four model variants were compared, along with a nine-variant ablation. The preprint was uploaded to arXiv on August 23, 2026, as 2608.22475.
The research team includes Diana Legziel Levy, Menachem Finkelstein, Peter Chin, Eilon Vaadia, and Sarel Cohen. Experiments used noiseless statevector simulation on 4 qubits, a regime chosen to reflect practical constraints of near-term superconducting hardware. The backbone-gradient variant improved accuracy in 3/4 seeds (mean +0.19%, 95% CI [-1.10%, +1.48%]) and consistently reduced Linear CKA similarity to baseline features (Δ=-0.025, 4/4 seeds), indicating structural reorganization of representations. The ablation identified simple shallow architectures as the most effective and reproducible configuration. Measurement-guided training consistently improved representation geometry without reducing accuracy. The paper explicitly claims no quantum computational advantage over classical methods.