Measurement-Induced Quantum Neural Networks

• MINN is a quantum machine learning architecture where measurements are integral to computation, enabling adaptive and nonlinear processing.
• Different implementations include adaptive monitored circuits, MBQC/cluster-state networks, qubit-neuron models, and learnable-readout variational models.
• Empirical studies show MINNs bridge classical and quantum regimes with optimized training methods and innovative measurement designs.

to=arxiv_search.search 天天中彩票追号json {"^^^^^^^^4query4 Quantum Neural Network\"^^^^^^^^4 OR ti:\4^^^^^^^^"Measurement-Induced Quantum Neural Network\"^^^^^^^^4 OR abs:\4^^^^^^^^"measurement-induced quantum neural network\"", "max_results": ^^^^4all:\4query4^^^^, "sort_by": "submittedDate", "sort_order": "descending"} to=arxiv_search.search 时时彩后रलjson {"^^^^^^^^4query4 OR id:(&&&^^^^4all:\4^^^^&&&) OR id:(&&&^^^^4 OR ti:\4^^^^&&&) OR id:(&&&^^^^4 OR abs:\4^^^^&&&) OR id:(Barney et al., 19 Mar 2025) OR id:(Chen et al., 10 Jan 2025) OR id:(Zhou et al., 2023)", "max_results": ^^^^4all:\4query4^^^^, "sort_by": "relevance", "sort_order": "descending"} Measurement-Induced Quantum Neural Network (MINN) denotes a family of quantum machine-learning architectures in which measurement is an internal computational primitive rather than a terminal readout. In this literature, measurement outcomes can determine later entangling gates, select adaptive measurement bases, generate layerwise activations, define stochastic feed-forward channels, or serve as trainable observables. The term therefore covers several non-identical constructions: adaptive monitored circuits, measurement-based or cluster-state quantum neural networks, layerwise qubit-neuron models driven by repeated measurements, and variational models with learnable readout operators (&&&^^^^4query4^^^^&&&, &&&^^^^4 OR ti:\4^^^^&&&, &&&^^^^4 OR abs:\4^^^^&&&, Barney et al., 19 Mar 2025, Chen et al., 10 Jan 2025).

^^^^4all:\4^^^^. Terminology and scope

The contemporary usage of MINN is plural rather than canonical. One line of work defines a MINN as a layered monitored circuit in which mid-circuit measurements feed forward into the parametrization of subsequent gates (&&&^^^^4query4^^^^&&&). A second line identifies MINNs with measurement-based quantum computation (MBQC) models, where a fixed graph or cluster state is consumed by adaptive local measurements whose angles are trainable (&&&^^^^4 OR abs:\4^^^^&&&, &&&^^^^4 OR ti:\4^^^^&&&). A third line realizes neurons directly as qubits, with each hidden layer implemented by qubit preparation, rotation, and measurement, so that measurement outcomes become the activations passed to later layers (Barney et al., 19 Mar 2025, Zhou et al., 2023). A fourth line uses the term for variational quantum models in which the measurement operator itself is optimized jointly with the circuit (Chen et al., 10 Jan 2025).

Family	Measurement role	Representative papers
Adaptive monitored-circuit MINN	Mid-circuit outcomes set later gate parameters	(&&&^^^^4query4^^^^&&&)
MBQC / cluster-state MINN	Fixed entangled resource is processed by tunable local measurements	(&&&^^^^4 OR abs:\4^^^^&&&, &&&^^^^4 OR ti:\4^^^^&&&)
Layerwise qubit-neuron MINN	Each layer’s activations are produced by projective or weak measurements	(Barney et al., 19 Mar 2025, Zhou et al., 2023)
Learnable-readout MINN	The observable or Hermitian readout is trainable	(Chen et al., 10 Jan 2025)

This multiplicity is not merely terminological. It reflects different answers to the same design question: where, in a quantum learning pipeline, should measurement back-action be allowed to participate in computation? In some models it alters future quantum dynamics; in others it provides the nonlinear hidden-layer interface; in others it determines the readout geometry. A related but distinct usage places the neural network on the classical side, as a decoder of measurement trajectories in monitored many-body systems rather than as the quantum model itself (&&&^^^^4 OR ti:\44^^^^&&&, &&&^^^^4 OR ti:\4^^^^5&&&).

^^^^4 OR ti:\4^^^^. Architectural realizations

In the adaptive monitored-circuit construction, a MINN is a layered ^^^^4all:\4^^^^D brick-wall circuit on PRESERVED_PLACEHOLDER_^^^^4query4^^^^ qubits with PRESERVED_PLACEHOLDER_^^^^4all:\4^^^^ layers. After each layer, a subset PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4^^^^ of qubits is measured in the computational basis, producing a register PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4^^^^, where $\mu_j^t=\pm1$ for measured sites and $0$ otherwise. The next layer’s gate parameters are generated by the nonlinear map

$$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$

with

$$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$

The paper studies a matchgate-restricted instantiation, generated by

$$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$

because this admits exact fermionic simulation, while a generic MINN is not expected to be efficiently classically simulable (&&&^^^^4query4^^^^&&&).

In MBQC realizations, the quantum state is prepared once as a graph state

$$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$

and computation is performed by single-qubit measurements such as

PRESERVED_PLACEHOLDER_^^^^4all:\4query4^^^^

The trainable object is then a measurement pattern PRESERVED_PLACEHOLDER_^^^^4all:\4all:\4^^^^ on a graph PRESERVED_PLACEHOLDER_^^^^4all:\4 OR ti:\4^^^^, inducing a channel PRESERVED_PLACEHOLDER_^^^^4all:\4 OR abs:\4^^^^. Determinism is enforced by flow: if a qubit measurement returns PRESERVED_PLACEHOLDER_^^^^4all:\44^^^^, compensation can be implemented by

PRESERVED_PLACEHOLDER_^^^^4all:\4^^^^5

or equivalently by adapting future measurement angles (&&&^^^^4 OR abs:\4^^^^&&&).

Two MBQC architectures are especially representative. The multiple-triangle ansatz (MuTA) builds a structured graph state out of five-qubit wires and triangle gadgets, with center qubits PRESERVED_PLACEHOLDER_^^^^4all:\4^^^^6 acting as tunable entangling controls; measuring them at PRESERVED_PLACEHOLDER_^^^^4all:\4^^^^7 yields non-entangling interactions, while PRESERVED_PLACEHOLDER_^^^^4all:\4^^^^8 yields entangling transformations (&&&^^^^4 OR abs:\4^^^^&&&). The measurement-based QCNN construction gives an exact cluster-state realization of general QCNNs by composing PRESERVED_PLACEHOLDER_^^^^4all:\4^^^^9-clusters for two-qubit filters and PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4query4^^^^-clusters for pooling, and it also demonstrates square-lattice cluster realizations in which all sites are measured in parameterized bases of the form PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4all:\4^^^^ (&&&^^^^4 OR ti:\4^^^^&&&).

A distinct architectural family quantizes classical feed-forward networks more directly. In the “natural quantization” model, each hidden-layer neuron is a single qubit initialized to PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4 OR ti:\4^^^^, rotated by

PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4 OR abs:\4^^^^

and then measured in the computational basis with PRESERVED_PLACEHOLDER_^^^^4 OR ti:\44^^^^, PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4^^^^5. The activation variable is defined by the outcome mapping PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4^^^^6 if PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4^^^^7 and PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4^^^^8 if PRESERVED_PLACEHOLDER_^^^^4 OR ti:\4^^^^9. Rotation angles depend explicitly on previous measurement outcomes, so the forward pass alternates quantum rotations and classical feed-forward (Barney et al., 19 Mar 2025). The “soft quantum feedforward neural network” realizes a related mechanism using classically controlled single-qubit operations and single-qubit measurements, with one qubit per neuron and no operational entangling gates during inference (Zhou et al., 2023).

Finally, in learnable-readout models, the circuit and encoding may be conventional, but the output observable is not fixed. Instead one learns a Hermitian PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4query4^^^^ and evaluates

PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4all:\4^^^^

This repositions the measurement stage as a trainable component of the hypothesis class rather than a static interface (Chen et al., 10 Jan 2025).

^^^^4 OR abs:\4^^^^. Measurement-induced dynamics and nonlinearity

The defining mechanism of a MINN is that measurement outcomes alter the effective computation. In adaptive monitored circuits, if PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4 OR ti:\4^^^^ denotes the full measurement history, then the trajectory probability is

PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4 OR abs:\4^^^^

and the expected objective is PRESERVED_PLACEHOLDER_^^^^4 OR abs:\44^^^^. Because later unitaries depend on earlier outcomes, measurement back-action produces genuinely history-dependent dynamics rather than a fixed unitary channel followed by sampling (&&&^^^^4query4^^^^&&&).

In MBQC, the same principle appears in a different form. The entangled resource is static, but adaptive local measurements and byproduct propagation supply effective nonlinearity and conditional branching. MuTA makes this explicit: stochastic measurement outcomes, together with classical feed-forward and the entangling role of the triangle-center qubits, provide the network-like behavior, while universality follows from the ability to implement arbitrary single-qubit rotations plus PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4^^^^5 (&&&^^^^4 OR abs:\4^^^^&&&). The measurement-based QCNN construction likewise replaces trainable coherent layers by trainable measurement bases on a pre-entangled cluster, with convolution and pooling implemented as measurement sequences on gadgetized subgraphs or on a square lattice (&&&^^^^4 OR ti:\4^^^^&&&).

In layerwise qubit-neuron models, the measurement-induced effect is especially transparent. For the projective model, the state after a rotation is

PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4^^^^6

so PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4^^^^7 and PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4^^^^8; these outcomes are then fed into the next layer’s classical preactivations (Barney et al., 19 Mar 2025). In the weak-measurement ancilla variant, a neuron qubit is entangled with an ancilla by PRESERVED_PLACEHOLDER_^^^^4 OR abs:\4^^^^9, and ancilla measurement induces Kraus maps $\mu_j^t=\pm1$^^^^4query4^^^^ and POVM elements $\mu_j^t=\pm1$^^^^4all:\4^^^^ that interpolate smoothly between no information at $\mu_j^t=\pm1$^^^^4 OR ti:\4^^^^ and projective readout at $\mu_j^t=\pm1$^^^^4 OR abs:\4^^^^ (Barney et al., 19 Mar 2025). The “soft quantum” architecture makes the same point in channel language: even when the underlying updates are linear CPTP maps, stochastic branching on measurement outcomes and classical conditioning generate a nonlinear dependence of expected outputs on inputs (Zhou et al., 2023).

A recurrent theme is the existence of controlled classical limits. In the rotation-activation model, $\mu_j^t=\pm1$4 recovers the exact classical binarized multilayer perceptron; in the weak-measurement variant, $\mu_j^t=\pm1$5 recovers deterministic classical activations (Barney et al., 19 Mar 2025). This suggests that many MINNs are best understood not as replacements for classical feed-forward structure but as measurement-driven deformations of it.

4. Readout theory, Fisher information, and measurement design

A central analytical result for MINNs concerns the statistics of restricted-support measurements. In regression on quantum states, the predicted label is

$\mu_j^t=\pm1$6

where $\mu_j^t=\pm1$7 acts nontrivially only on $\mu_j^t=\pm1$8 measured qubits. If $\mu_j^t=\pm1$9, then the prediction variance depends on both the spectrum $0$^^^^4query4^^^^ and the degeneracy structure of the projectors $0$^^^^4all:\4^^^^. The paper shows that measurements with restricted support generally have fewer effective outcome bins and therefore lower classical Fisher information $0$^^^^4 OR ti:\4^^^^, leading to the bound

$0$^^^^4 OR abs:\4^^^^

where $0$4 is the quantum Fisher information (&&&^^^^4all:\4^^^^&&&).

For the convex-mixture regression model, the optimal restricted-support variance takes the explicit form

$0$5

which is independent of the mixture parameter $0$6 and decreases monotonically with the number $0$7 of measured qubits. The same analysis gives concrete examples: for $0$8, $0$9; for $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$^^^^4query4^^^^, $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$^^^^4all:\4^^^^. The paper interprets this as a spectral-resolution effect: restricted support compresses the outcome space, increases degeneracy, lowers $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$^^^^4 OR ti:\4^^^^, and inflates prediction variance (&&&^^^^4all:\4^^^^&&&).

This result matters directly for QCNN-like MINNs, where pooling and tracing out qubits are architectural primitives. The analysis indicates that aggressive support restriction at readout can raise shot complexity and degrade trainability by making label prediction intrinsically noisier. The paper therefore recommends increasing readout support, engineering observables with more distinct eigenvalues and lower degeneracy, using ancillas and Naimark extensions to implement finer POVMs, and including a variance penalty in the training objective: $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$^^^^4 OR abs:\4^^^^ These conclusions are deterministic and do not rely on Haar-randomness or unitary-design assumptions (&&&^^^^4all:\4^^^^&&&).

The same section of the literature also identifies a qualified exception to the heuristic that “more measured qubits are better.” For pure-state families lying in a real two-dimensional subspace, there exists an observable $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$4 such that

$$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$5

so even a small-support measurement can saturate both the classical and quantum Cramér–Rao limits when the basis is properly aligned (&&&^^^^4all:\4^^^^&&&). This exception complements learnable-readout approaches, where the motivation for training $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$6 is precisely to align the readout with the data manifold and to expand or reposition the output range via the observable’s spectrum (Chen et al., 10 Jan 2025).

5. Training procedures and empirical demonstrations

Training methods vary with the role played by measurement. Adaptive monitored-circuit MINNs use score-function estimators,

$$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$7

with an optimal baseline $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$8 to reduce variance (&&&^^^^4query4^^^^&&&). Variational and MBQC models often use parameter-shift for gate angles and SGD, Adam, or RMSProp for classical optimization (&&&^^^^4 OR ti:\4^^^^&&&, &&&^^^^4 OR abs:\4^^^^&&&, Chen et al., 10 Jan 2025). The layerwise neuron models use stochastic gradient descent with momentum and a clipped straight-through estimator for the non-differentiable binarized activation, while the photonic GKP-constrained MuTA study supplements gradient methods with $$\theta^{t+1} = \frac{\pi}{2}\left[1 - \phi_a(W^t \mu^t + b^t)\right],$$9-greedy search and Deep Q-Networks for discrete angle selection (Barney et al., 19 Mar 2025, &&&^^^^4 OR abs:\4^^^^&&&).

Task	Architecture	Reported outcome
QFI-state classification	MuTA MBQC QNN	Accuracy $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$^^^^4query4^^^^ (&&&^^^^4 OR abs:\4^^^^&&&)
Instrument learning	MuTA teleportation	Perfect fidelity from qubit 8 to ^^^^4 OR ti:\4 OR ti:\4^^^^ (&&&^^^^4 OR abs:\4^^^^&&&)
MNIST subset	Rotation + weak-measurement MINN	Best validation error $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$^^^^4all:\4^^^^ at $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$^^^^4 OR ti:\4^^^^, $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$^^^^4 OR abs:\4^^^^ (Barney et al., 19 Mar 2025)
Speaker recognition	VQC + learnable Hermitian readout	$$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$4 test accuracy with separate optimizers (Chen et al., 10 Jan 2025)
Handwritten digits $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$5	Soft quantum feedforward network	$$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$6 accuracy (Zhou et al., 2023)

The MBQC literature supplies several controlled demonstrations. MuTA learns single-qubit Haar gates and $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$7 using $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$8, $$\phi_a(x)=\mathrm{htanh}(x/a)= \begin{cases} x/a, & |x|\le a,\ \mathrm{sgn}(x), & |x|>a. \end{cases}$$9, Adam, and averages over ^^^^4 OR ti:\4query4^^^^ random initializations and target unitaries; it also achieves perfect-fidelity teleportation after training on ^^^^4 OR abs:\4^^^^5 states and testing on ^^^^4all:\4^^^^5, averaged over ^^^^4all:\4query4^^^^ runs (&&&^^^^4 OR abs:\4^^^^&&&). The measurement-based QCNN on square-lattice clusters reproduces Haldane phase boundaries and, on Iris classification with ^^^^4 OR ti:\4^^^^8 parameters matched across baselines, converges as fast as gate-based QCNN and faster than a classical CNN, with higher final test accuracy than QCNN in that setup (&&&^^^^4 OR ti:\4^^^^&&&).

The neuron-based MINN literature emphasizes smooth interpolation between classical and quantum regimes. In the natural-quantization model with 5,^^^^4query4query4query4^^^^ training images and ^^^^4all:\4query4^^^^,^^^^4query4query4query4^^^^ validation images from MNIST, the best validation error for pure rotation-activation quantization is $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$^^^^4query4^^^^ at $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$^^^^4all:\4^^^^, the best pure weak-measurement result is $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$^^^^4 OR ti:\4^^^^ at $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$^^^^4 OR abs:\4^^^^, and the combined model reaches $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$4 at $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$5, $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$6. The same work reports a sharp loss of learnability below approximately $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$7 (Barney et al., 19 Mar 2025). The soft quantum feedforward model reports $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$8 test accuracy on XOR after the first epoch, $$H = \frac{1}{2}(\theta_{XX} X\otimes X + \theta_{XY} X\otimes Y + \theta_{YX} Y\otimes X + \theta_{YY} Y\otimes Y + \theta_{ZI} Z\otimes I + \theta_{IZ} I\otimes Z),$$9 on circles, $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$^^^^4query4^^^^ on moons, and $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$^^^^4all:\4^^^^ on the $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$^^^^4 OR ti:\4^^^^ handwritten-digit task, outperforming QuantumFlow by $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$^^^^4 OR abs:\4^^^^ and a PQC baseline by $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$4 in that last setting (Zhou et al., 2023).

Readout learning also yields substantial empirical gains. On make_moons with 4 qubits and ^^^^4 OR ti:\4^^^^ variational layers, a VQC with learnable observable outperforms fixed-measurement VQC across noise levels, and on a ^^^^4all:\4query4^^^^-speaker VCTK subset the reported test accuracies are $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$5 for fixed Pauli-$$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$6, $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$7 for a learnable Hermitian observable, and $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$8 when separate optimizers with different learning rates are used for unitary and Hermitian parameters (Chen et al., 10 Jan 2025). The adaptive matchgate MINN, finally, is applied to continuous optimization, MNIST classification, and Sherrington–Kirkpatrick ground-state search; the reported behavior is smooth loss reduction and effective training over a broad range of monitoring rates, with best-sample energies approaching the global or exact ground-state minima in the studied instances (&&&^^^^4query4^^^^&&&).

6. Limitations, adjacent usages, and open problems

Several misconceptions recur in discussions of MINNs. First, MINN is not a single architecture class. The term encompasses at least adaptive monitored circuits, MBQC/cluster-state networks, qubit-neuron feed-forward models, and learnable-readout variational models; these share a measurement-centric computational role but differ substantially in state preparation, adaptivity, and trainability (&&&^^^^4query4^^^^&&&, &&&^^^^4 OR abs:\4^^^^&&&, Barney et al., 19 Mar 2025, Chen et al., 10 Jan 2025). Second, MINN is not synonymous with MBQC. MBQC is one important realization, but repeated projective or weak measurement with classical feed-forward also defines MINNs in the neuron-based and monitored-circuit literatures (&&&^^^^4 OR ti:\4^^^^&&&, Barney et al., 19 Mar 2025). Third, local readout is not universally suboptimal: the restricted-support variance penalty is generic in the regression setting analyzed in (&&&^^^^4all:\4^^^^&&&), but real two-dimensional pure-state subspaces furnish explicit saturating exceptions.

The principal technical limitations are likewise architecture-dependent. Adaptive monitored circuits require low-latency mid-circuit readout and fast classical feed-forward, and their score-function gradients can have high variance (&&&^^^^4query4^^^^&&&). MBQC models require scalable graph-state preparation and classical control for adaptivity; under photonic GKP constraints, the accessible measurement angles are restricted to $$\ket{G} = \prod_{(i,j)\in E} CZ_{ij}\ket{+}^{\otimes n},$$9, with PRESERVED_PLACEHOLDER_^^^^4all:\4query4query4^^^^ requiring PRESERVED_PLACEHOLDER_^^^^4all:\4query4all:\4^^^^ magic-state injection, and the discrete optimization problem is NP-hard (&&&^^^^4 OR abs:\4^^^^&&&). Learnable-observable models face measurement overhead because a full Hermitian readout scales as PRESERVED_PLACEHOLDER_^^^^4all:\4query4 OR ti:\4^^^^ in a Pauli expansion, which motivates sparse or low-parameter parameterizations (Chen et al., 10 Jan 2025). QCNN-like architectures face a separate readout-design problem: aggressive pooling and restricted final support can increase prediction variance even when they improve circuit depth or avoid barren plateaus (&&&^^^^4all:\4^^^^&&&).

A nearby but distinct research direction uses neural networks to decode the effects of many measurements in monitored many-body systems. In one approach, an unsupervised classical model learns the map PRESERVED_PLACEHOLDER_^^^^4all:\4query4 OR abs:\4^^^^ from extensive measurement records to post-measurement probe states, with cross-correlation estimators certifying measurement-induced entanglement and revealing a learnability transition aligned with a measurement-induced phase transition (&&&^^^^4 OR ti:\44^^^^&&&). In another, neural decoders map spacetime measurement records to the state of a reference qubit in hybrid monitored circuits, and the learnability of that decoder itself exhibits sharp behavior across the measurement-induced entanglement transition (&&&^^^^4 OR ti:\4^^^^5&&&). These works do not define the neural network as the quantum model, but they show that measurement-induced learning can also mean learning the structure generated by monitored dynamics rather than learning with a monitored quantum model.

Open problems follow directly from this fragmented landscape. The MBQC literature identifies the absence of general conditions for quantum advantage over gate-based QML, the need for richer kernel design, and the challenge of hardware-constrained training algorithms (&&&^^^^4 OR abs:\4^^^^&&&). The learnable-readout literature leaves open principled priors for PRESERVED_PLACEHOLDER_^^^^4all:\4query44^^^^, automated sparsification of measurement terms, and extensions to learned POVMs (Chen et al., 10 Jan 2025). The monitored-circuit literature highlights the need for lower-variance gradient estimators and for a systematic theory of how monitoring rate, entanglement generation, and trainability interact beyond exactly simulable matchgate cases (&&&^^^^4query4^^^^&&&). Across all of these strands, the central unresolved issue is not whether measurement can be made part of a neural architecture, but how measurement back-action, spectral design, and classical feed-forward should be co-designed so that measurement becomes an asset rather than a bottleneck.