{"id":"arxiv/2602.06699","url":"https://arxiv.org/abs/2602.06699","tags":["quant-ph","cs.CL","cs.LG"],"score":0,"title":"Quantum Attention by Overlap Interference: Predicting Classical and Many-Body Quantum Sequences","domain":"quant-ph","source":"arxiv","authors":["Alessio Pecilli","Matteo Rosati"],"reviews":0,"subject":"quant-ph","summary":"We propose a variational quantum implementation of self-attention (QSA)-the core operation in transformers and large language models-which predicts future elements of a sequence by forming overlap-weighted combinations of past data. At variance with previous approaches, our QSA realizes the required nonlinearity through interference of state overlaps and a degree-$k$ polynomial kernel, and estimates a loss based on Rényi-$1/2$ entropic functionals via two observables' expectation values, avoiding the decoding of amplitude-encoded predictions into classical probabilities. QSA also…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-02-06","reactions":{}}