---
title: Conformalized Quantum DeepONet Ensembles
url: https://www.emergentmind.com/topics/conformalized-quantum-deeponet-ensembles
type: topic
---

# Conformalized Quantum DeepONet Ensembles

Conformalized Quantum DeepONet Ensembles are a quantum machine learning framework for operator learning that combines Quantum Orthogonal Neural Networks (QOrthoNNs), ensemble-based epistemic modeling, adaptive conformal prediction, and Superposed Parameterized Quantum Circuits (SPQCs) in order to address two limitations identified for existing approaches: quadratic inference complexity and unreliable uncertainty quantification in safety-critical settings [2605.00330]. In the formulation reported in the literature, the framework targets scalable evaluation over fine discretizations by reducing operator inference complexity from $\mathcal O(n^2)$ to $\mathcal O(n)$, while also providing distribution-free coverage guarantees through conformal prediction. The reported empirical setting includes synthetic partial differential equations and real-world power system dynamics, with uncertainty behavior examined under realistic quantum noise [2605.00330].

## 1. Operator-learning setting and the DeepONet baseline

The underlying task is operator learning, in which an operator
$$
\mathcal G: u(\cdot)\mapsto y = \mathcal G[u](\cdot)
$$
is approximated by a DeepONet architecture. In the stated construction, DeepONet separates the computation into a branch network and a trunk network,
$$
\mathbf b = \mathrm{Branch}_\theta(u)\in\mathbb R^p, \qquad
\mathbf t = \mathrm{Trunk}_\theta(y)\in\mathbb R^p,
$$
followed by a dot-product readout,
$$
\mathcal G_\theta[u](y)\;=\;\mathbf b^\top\mathbf t.
$$

The scaling bottleneck arises from the dense layers used in the branch or trunk components. For a network of hidden width $n$, each dense layer costs $\mathcal O(n^2)$. When the trunk must be evaluated $M$ times for $M$ query points, the per-input cost becomes $\mathcal O(M\,n^2)$. The source text explicitly identifies this as prohibitive for large $M$ and for extensive Monte-Carlo runs in the range $10^4$–$10^5$ scenarios [2605.00330].

The motivation for a quantum implementation is stated in three parts. First, parameterized quantum circuits can in principle emulate orthogonal transforms in sub-quadratic depth. Second, quantum uncertainty in the form of shot noise does not by itself capture epistemic model uncertainty. Third, naive quantum ensembling multiplies qubit requirements by the ensemble size $L$. This combination motivates a joint quantum-plus-conformal design rather than a purely quantum surrogate or a purely classical uncertainty wrapper.

## 2. QOrthoNNs and sub-quadratic operator inference

QOrthoNNs are introduced as the mechanism for replacing a classical orthogonal layer
$$
\mathbf y = W\,\mathbf x,\quad
W\in\mathbb R^{n\times n},\quad
W^\top W = I_n
$$
with a quantum circuit implementation [2605.00330]. The reported construction uses $n$ data qubits $\{q_1,\ldots,q_n\}$ plus one ancilla and consists of three stages:

1. $S(\mathbf x)$, an amplitude-loader into the unary subspace.
2. $U(\boldsymbol\theta)$, a pyramidal RBS (Reconfigurable Beam Splitter) network implementing $W$.
3. Tomography plus ancilla manipulation to read out each $y_j$.

The complexity claim is central. Classically, each orthogonal layer remains $\mathcal O(n^2)$. In the quantum realization, the data loader depth is stated as $\sim n$, the pyramidal unitary depth as $\sim 2n$, and tomography as $\sim 2n$ using two loaders plus $\mathcal O(1)$ gates. The resulting total depth is therefore $\mathcal O(n)$. Because amplitude encoding and tomography each require $\tilde{\mathcal O}(1/\delta^2)$ shots for error $\delta$, the end-to-end cost is
$$
\tilde{\mathcal O}\bigl(n/\delta^2\bigr),
$$
which is described as sub-quadratic in $n$ [2605.00330].

This reduction is significant specifically for operator-learning workloads in which fine discretizations force repeated trunk evaluations. A plausible implication is that the advantage is most relevant when the cost of repeated classical dense-layer application dominates the overall surrogate evaluation budget.

## 3. SPQCs and ensemble compression

The framework does not rely on a single QOrthoNN model. It uses ensembles to represent epistemic uncertainty, but it also addresses the hardware cost of quantum ensembling. The baseline comparison is explicit: $L$ independent QOrthoNN models in parallel require $L\times(n+1)$ qubits, whereas an SPQC encodes all $L$ models in one circuit using $\log_2L$ address qubits [2605.00330].

The SPQC state preparation is given by
$$
|\Psi\rangle
= \frac1{\sqrt L}
\sum_{j=1}^L
|j_{\rm addr}\rangle
\otimes
U\bigl(\boldsymbol\theta^{(j)}\bigr)\,
S\bigl(\mathbf x^{(j)}\bigr)\,
|0\rangle^{\otimes n}.
$$
In this construction, the address register $|j_{\rm addr}\rangle$ selects model $j$, and controlled unitaries conditionally load data $\mathbf x^{(j)}$ and apply $U(\boldsymbol\theta^{(j)})$.

The extraction mechanism is also specified: after a single joint measurement, with post-selection to unary outcomes, all $L$ model outputs are obtained simultaneously. The circuit depth scales as $\mathcal O(L\,n)$ because of controlled-RBS ladders, while the qubit count is only
$$
n+1+\lceil\log_2L\rceil.
$$

The framework therefore trades linear qubit growth for logarithmic address overhead at the cost of circuit-depth growth in $L$. This trade-off is later reflected in the stated limitations: SPQC depth $\mathcal O(L\,n)$ can accumulate noise for large $L$. A common misconception addressed directly by this design is that quantum parallelism alone makes ensembling effectively free; in the reported formulation, naive parallelism still scales hardware resources linearly with the number of models.

## 4. Conformal prediction and distribution-free uncertainty

The uncertainty layer combines ensemble statistics with adaptive conformal prediction [2605.00330]. For each calibration point $(u_i,y_{ij},s_{ij})$, the ensemble mean $\mu_{ij}$ and standard deviation $\sigma_{ij}$ are computed, and the adaptive nonconformity score is defined as
$$
r_{ij}\;=\;
\frac{\bigl|s_{ij}-\mu_{ij}\bigr|}{\sigma_{ij}+\varepsilon},
\quad \varepsilon\ll1.
$$

Let $\hat q$ denote the $(1-\alpha)$-quantile of the calibration scores $\{r_{ij}\}$. For a new input $(u,y)$, the prediction set is
$$
C_\alpha(u,y)
\;=\;
\bigl\{v\in\mathbb R:\,|v-\mu(u,y)|\le \hat q\,\sigma(u,y)\bigr\}.
$$
Under the standard conformal assumption of exchangeability between calibration and test data, the framework has the finite-sample guarantee
$$
\Pr\bigl[s\in C_\alpha(u,y)\bigr]\;\ge\;1-\alpha.
$$

The conceptual role of conformalization is explicit. Quantum uncertainty arising from shot noise alone is not treated as a sufficient account of model uncertainty, so the framework retains model ensembles and then calibrates their outputs with conformal prediction. This separates epistemic modeling from finite-sample calibration. The reported guarantee is distribution-free, but only under exchangeability; the same source notes that this assumption is violated in streaming and temporal tasks, which is an important qualification rather than a minor technical caveat.

## 5. Reported empirical results

The reported experiments cover synthetic PDE benchmarks and real-world power system benchmarks, with a target coverage of $90\%$ throughout [2605.00330].

For the synthetic PDE setting, two benchmark operators are listed: the antiderivative $\mathcal F: v\mapsto u$ with $\mathrm d u/\mathrm d x=v$, $u(0)=0$, and the one-dimensional advection equation $\partial_x u+\partial_t u=0$ with periodic boundary conditions.

The ideal simulation results are:

- **Antiderivative, $L=4$**: Relative $L_2$ error $0.46\%$, coverage $88.4\%$, average width $0.004$, peak uncertainty $\sigma=0.044$.
- **Antiderivative, $L=8$**: Relative $L_2$ error $0.46\%$, coverage $92.1\%$, average width $0.005$, peak uncertainty $\sigma=0.080$.
- **Advection, $L=4$**: Relative $L_2$ error $2.38\%$, coverage $89.4\%$, average width $0.062$, peak uncertainty $\sigma=0.751$.
- **Advection, $L=8$**: Relative $L_2$ error $2.28\%$, coverage $89.0\%$, average width $0.053$, peak uncertainty $\sigma=0.621$.

For the power system setting, three tasks are listed: offline voltage-to-voltage forecasting $(\mathcal H)$ from $1.9\,\mathrm s$ pre-fault to $2.0\,\mathrm s$ post-fault, offline voltage-to-active-power $(\mathcal F)$, and online sliding-window voltage-to-voltage $(\mathcal G)$.

The ideal simulation results are:

- **Online V→V**: Relative $L_2$ error $5.49\%$, coverage $90.1\%$, average width $0.148$, peak uncertainty $\sigma=1.328$.
- **Offline V→V**: Relative $L_2$ error $12.49\%$, coverage $89.7\%$, average width $0.345$, peak uncertainty $\sigma=2.691$.
- **Offline V→P**: Relative $L_2$ error $4.08\%$, coverage $89.7\%$, average width $0.001$, peak uncertainty $\sigma=0.063$.

Taken together, these experiments are presented as demonstrating accurate predictions with calibrated uncertainty. This suggests that the conformal layer remains effective across both synthetic operator benchmarks and application-driven dynamical-system surrogates, although the reported accuracy and interval widths vary substantially by task.

## 6. Quantum noise, limitations, and stated future directions

The impact of quantum noise is explicitly examined under hardware-calibrated depolarizing and readout noise from IBM Marrakesh, Torino, and Brisbane [2605.00330]. The reported finding is that empirical coverage remains at least $90\%$ across shot budgets from $5\,\mathrm k$ to $100\,\mathrm k$. The same source additionally reports that, for very shallow circuits with at most $7$ qubits, increased depolarizing noise can act as a regularizer, slightly improving coverage and sharpening intervals.

The strengths claimed for the framework are equally explicit: sub-quadratic inference $\tilde{\mathcal O}(n)$ via QOrthoNN, provable distribution-free uncertainty via conformal prediction, and logarithmic-scale qubit cost in ensembles via SPQC. The limitations are also directly enumerated. Conformal validity assumes calibration/test exchangeability and is violated in streaming or temporal tasks; near-term quantum devices remain noisy, so classical-quantum hybrid variants may be needed; and SPQC depth $\mathcal O(L\,n)$ can accumulate noise for large ensemble sizes.

The future directions listed in the source are: adaptive conformal methods for non-exchangeable data, including weighted and Mondrian variants; theoretical study of noise-induced regularization in shallow PQCs; hardware demonstrations on more than $20$ qubits as coherence improves; and multi-fidelity hybrid architectures combining a classical branch, a quantum trunk, and an SPQC ensemble. A plausible implication is that subsequent work will focus less on the existence of a conformalized quantum operator-learning pipeline and more on robustness under temporal shift, hardware noise accumulation, and hybrid architectural decomposition.

Source: https://www.emergentmind.com/topics/conformalized-quantum-deeponet-ensembles