Theory and interpretability of Quantum Extreme Learning Machines: a Pauli-transfer matrix approach

TL;DR

This work develops a Pauli-transfer matrix (PTM) framework to analyze quantum extreme learning machines (QELMs), linking data encoding, reservoir mixing, and measurement to a classical feature-decoding problem. By expanding states in the Pauli basis and describing channels with PTMs, it derives a classical surrogate model f(x) = w^T R phi(x) and introduces decodability scores gamma_r^2 = (R^+R)_{rr} and nonlinear capacity metrics R^2(k). The analysis combines Krylov growth and Pauli weight spreading to explain how operator spreading and measurement subsets constrain expressivity, and shows that temporal multiplexing can enlarge the decodable feature subspace. The authors illustrate the framework with time-series modeling of chaotic systems (Lorenz-63 and Halvorsen), producing interpretable flow-map surrogates and offering design guidance for QELMs on NISQ hardware.

Abstract

Quantum reservoir computers (QRCs) have emerged as a promising approach to quantum machine learning, since they utilize the natural dynamics of quantum systems for data processing and are simple to train. Here, we consider n-qubit quantum extreme learning machines (QELMs) with continuous-time reservoir dynamics. QELMs are memoryless QRCs capable of various ML tasks, including image classification and time series forecasting. We apply the Pauli transfer matrix (PTM) formalism to theoretically analyze the influence of encoding, reservoir dynamics, and measurement operations, including temporal multiplexing, on the QELM performance. This formalism makes explicit that the encoding determines the complete set of (nonlinear) features available to the QELM, while the quantum channels linearly transform these features before they are probed by the chosen measurement operators. Optimizing a QELM can therefore be cast as a decoding problem in which one shapes the channel-induced transformations such that task-relevant features become available to the regressor. The PTM formalism allows one to identify the classical representation of a QELM and thereby guide its design towards a given training objective. As a specific application, we focus on learning nonlinear dynamical systems and show that a QELM trained on such trajectories learns a surrogate-approximation to the underlying flow map.

Figures (13)

• Figure 1: Time evolution of the Pauli weight average $\bar{\nu}_k$ [\ref{['eq_pauli_wt_avg']}] of a single-site Pauli $\sigma_k$ in the $h$--$J$ coupling parameter space of the TFIM \ref{['eq_H_TFIM_zzx']} for $n=3$ qubits.
• Figure 2: Visualization of the operator spreading \ref{['eq_pauli_op_spread']} via the effective PTM $R$ [\ref{['eq_tmux']}] constructed in the temporal multiplexing scheme for the TFIM (a) and random unitary (b). The selected observable set $\mathcal{S}$ are all weight-1 Pauli operators of a $n=3$ qubit system. With increasing evolution time (from top to bottom), these initial operators develop nonzero projection onto the full $d^2$-dimensional Pauli basis [\ref{['eq_pauli_op_spread']}]. In the TFIM case, the supporting basis exhibits a sparse pattern, whereas it is dense for a random unitary.
• Figure 3: Projector $R^+R$ of the effective PTM [\ref{['eq_tmux']}] for different unitary types and a complete weight-1 Pauli observable set. Panels (a,b) directly correspond to the $R$ shown in \ref{['fig:tmux_eff_ptm']}, which are constructed by temporal multiplexing. Panel (c) shows the projector based on $R=V(t=0.5)$ (no temporal multiplexing). The dashed white lines indicate features for which $\gamma_j^2>0.5$, while the black lines indicate Pauli weight sector boundaries. Due to operator spreading over a time $t=0.5$, decodability in (c) has significantly decreased.
• Figure 4: Feature decodability score $\bar{\gamma}_j^2$ obtained from the effective PTM $R$ [\ref{['eq_decod_score']}], averaged over each Pauli weight sector, of a $n=3$ qubit model as a function of temporal multiplexing iteration length ${L}$ [\ref{['eq_tmux']}]. The unitary evolution time is given by $t_{L}={L}$ (i.e., initial time $t_1=1$). The corresponding rank and number of rows of $R$, representing the measurement budget (${B} = {m} {L}$), are shown on the top axis (as a fraction of the total operator space dimension $d^2=64$). We consider the TFIM \ref{['eq_H_TFIM_zzx']} in (a,c,e), the TFIM \ref{['eq_H_TFIM_xxz']} in (g), and random Hamiltonians in (b,d,f). The observable set $\mathcal{S}$ is either only $Z$-Paulis (a,b), $Z$- and $ZZ$-Paulis (c,d,g), or all weight-1 Paulis (e,f).
• Figure 5: Ideal nonlinear capacity score $\mathcal{R}^2(k)$ [\ref{['eq_nlcap_R2_score']}] for amplitude and rotational encodings [\ref{['eq_enc_amplsqrt', 'eq_enc_rot']}], indicating the ability of an ideal QELM $f(\mathbf{u}) = \mathbf{w}^\top \boldsymbol{\upphi}(\mathbf{u})$ to generate degree-$k$ monomials of the input $u_\alpha$ ($\alpha=1,\ldots,n$). Unless indicated, only Pauli-$Z$ features are measured in (a). Nonlinear capacity is significantly enhanced when measuring over all Paulis (b). This is also demonstrated by the score for amplitude encoding with Pauli-$X$ readout included in (a). The dash-dotted curve represents the theoretical score for $\mathcal{R}^2_Z(k)$ with amplitude encoding and $n=5$ input qubits [\ref{['eq_nlcap_R2_theo_Z']}], which agrees exactly with the numerical results. $\Sigma$ denotes the integrated capacity over all $k$. The curves drop with increasing $k$ is because the quantum model can only generate square-free monomials of the Pauli features $\phi_j$.