271st RQC Seminar

• 講演者

  Mr. Alberto Ferrara
  ( University of Palermo )

• 日程

  2026年4月27日(月), 16:00 - 17:00(4:00 p.m.-5:00 p.m.)

• 開催場所

  ハイブリッド（Zoom,
  345-347 Seminar Room, 3F, Main Research Building, Wako Campus / 和光地区 研究本館3階 セミナー室 (345-347) (C01))

• 講演タイトル

  From avoided level crossings to recurrent quantum circuits: Strategies for quantum information extraction

• お問合せ

  norilab_rqc_assist[at]ml.riken.jp

講演概要
In this presentation we will discuss two distinct but complementary quantum platforms for extracting information from complex dynamics: optomechanical readout of avoided level crossings, and feedback‑enhanced quantum reservoir computing.

In the first part, we propose a quantum state readout protocol for systems exhibiting an avoided level crossing, modeled as an effective two-level system coupled to standard optomechanical cavities. The radiation‑pressure interaction alters the eigenbasis such that the adiabatic eigenstates are no longer stationary, inducing Rabi‑like oscillations between the bare states and their superpositions. This coupling enables continuous, real‑time access to the quantum state via the cavity output field, providing a direct probe of the underlying dynamics. Notably, in the presence of noise, the leaky cavities drive the system toward a steady state that stabilizes the internal oscillations—a counterintuitive effect that can be harnessed for state preservation or readout synchronization. We illustrate the robustness of this approach by focusing on optomechanical two‑photon hopping, a specific Casimir‑type process, and demonstrate that clear signatures of the quantum dynamics persist in the readout signal even under thermal noise.

In the second part, we propose a quantum reservoir computing protocol based on a cascaded quantum circuit with Haar-random unitary operations and a set of memory qubits enabling quantum feedback at each step of the information processing. Input states, either uncorrelated or entangled, are sequentially injected and processed, while projective measurements at each step produce a classical probability vector that serves as a feature space for a shallow neural network. The readout layer is subsequently trained to reconstruct both local and non‑local observables, including entanglement witnesses and non‑linear functions of the input density matrices. We investigate the role of quantum memory in enhancing retrieval fidelity and demonstrate that the protocol can accurately estimate non linear properties such as state overlaps and non‑stabilizerness measures, even under finite measurement statistics.