Quantum Incremental Learning with Mixed State Prototypes

Kang Hao Cheong; Qianli Zhou; Wen Jiang; Witold Pedrycz; Xinyang Deng; Yu Wu

REVIEW 3 major objections 6 minor 59 references

Quantum Incremental Learning with Mixed State Prototypes

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A fixed-width quantum classifier can keep learning new classes by adding mixed-state prototypes instead of widening its circuit.

desk verdict A genuinely new combination—mixed-state prototypes for class-incremental learning on a fixed-width quantum circuit—with correct math and honest experiments, but the NISQ-era noise-robustness claim is untested because the simulations are noiseless. read the letter →

arxiv 2608.10464 v1 pith:CT2LSIOY submitted 2026-08-11 cs.AI cs.LG

Yu Wu , Qianli Zhou , Xinyang Deng , Wen Jiang , Kang Hao Cheong , Witold Pedrycz This is my paper · ORCID

classification cs.AIcs.LG

keywords quantumneuralnetworkincrementallearningmetricmixedstatescontinualmixed-stateprototypesHilbert-SchmidtdistanceSWAPtest

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Class-incremental learning usually forces a model to either store old data, freeze old weights, or expand its architecture; for quantum classifiers, expanding the circuit width is especially disruptive because it changes the Hilbert space and erases learned entanglement. This paper argues that a quantum classifier can instead keep a fixed-width shared backbone and accommodate every new class by adding one small trainable prototype circuit that represents the class as a mixed quantum state. The claim is that mixed-state prototypes beat pure-state prototypes at capturing intra-class variation, and that classification by Hilbert–Schmidt distance can be computed with cheap weighted SWAP tests rather than growing measurement heads. If the claim holds, adding a category costs a fixed 180-parameter 4-qubit circuit plus a handful of mixture weights, while the 8-qubit backbone never widens. Simulations on CIFAR-100 splits and TinyImageNet show the classifier outperforms direct-measurement quantum baselines and stays competitive with classical prototype methods.

What carries the argument

The load-bearing object is the convex-combination-of-pure-states (CCPS) ansatz, which represents a rank-$K$ mixed-state prototype as $\rho_c = \sum_{i=0}^{K-1} w_{c,i} U_c |i\rangle\langle i| U_c^\dagger$ with softmax-parameterized weights $w_{c,i} = \exp(s_{c,i})/\sum_j \exp(s_{c,j})$. The companion identity is the decomposition of the squared Hilbert–Schmidt distance into a classical purity term and a weighted sum of SWAP-test overlaps: $\|\rho-\sigma_{\mathrm{CCPS}}\|_F^2 = \mathrm{Tr}[\rho^2] + \sum_i p_i^2 - 2\sum_i p_i F_i$, where each $F_i = 2P(0)-1$ comes from measuring an ancilla in a SWAP test. This decomposition is what lets the framework classify without expanding the measurement structure: the per-class logit is the weighted overlap minus half the prototype purity, and the shared 8-qubit backbone and its 4-qubit output space stay fixed while new classes only add prototype circuits with 180 parameters each.

What would settle it

Run Algorithm 1 on a noisy 4-qubit processor or with a small number of measurement shots, compute the estimated logits, and compare them with exact Hilbert–Schmidt distances from full-state simulation: if any query's predicted label changes because the SWAP-test error exceeds half the smallest margin between the top-two class logits, then the claimed NISQ-era viability fails.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a fixed-width quantum feature extractor can serve as a permanent backbone for class-incremental learning if each class is represented as a trainable mixed-state prototype rather than as an output head. Concretely, an input $x$ is mapped by an 8-qubit QCNN to a $16\times 16$ density matrix $\rho(x)$ on 4 retained wires, and each class $c$ is assigned a CCPS prototype $\rho_c = \sum_{i=0}^{K-1} w_{c,i} U_c |i\rangle\langle i| U_c^\dagger$; the prediction is $\hat y = \arg\min_c \|\rho(x)-\rho_c\|_F^2$. Because the query purity term is class-independent and the prototype purity is a classical sum of squared weights, the squared Hilbert–Schmidt distance reduces to a weighted sum of SWAP-test overlaps minus a precomputable offset, which is exactly the logit used in Algorithm 1. New classes are incorporated by fitting a new prototype against the frozen backbone and selecting nearest-to-prototype exemplars for replay, so the circuit width remains fixed. The paper reports that this mixed-state representation outperforms pure-state prototypes and direct probability-measurement quantum classifiers, and that the prototype rank $K$ acts as a PCA-like knob filtering low-contribution components.

Load-bearing premise

The whole distance-based classification scheme depends on the SWAP-test overlap estimates $F_i = 2P(0)-1$ staying accurate enough that the predicted nearest prototype matches the true Hilbert–Schmidt distance; the experiments use only noiseless simulation with no shot budget or error mitigation, so on real hardware measurement noise could bias these overlaps and reorder the class logits.

Editorial extensions

If this is right

If the central claim is correct, the number of classes can grow without ever widening the quantum circuit: each new class adds one 180-parameter 4-qubit prototype circuit and $K$ mixture weights, leaving the frozen 8-qubit backbone untouched.
Because the squared Hilbert–Schmidt distance decomposes into precomputable purities and weighted SWAP-test overlaps, inference logits are directly measurable by $K$ SWAP tests per class, with no expanded readout or additional observables.
The rank $K$ of each prototype acts as a low-rank approximation of the class density matrix, and the paper's rank sweep shows that mid-range $K$ values beat both $K=1$ (pure-state prototype) and the full-rank $K=16$ setting on several benchmarks.
On the simulated benchmarks, the mixed-state prototype classifier improves on direct-measurement quantum baselines by 12–16 percentage points on average and shows a gradual, not abrupt, accuracy decline across incremental stages.
The framework's parameter complexity is $O(n\log_2(\max(n,d)))$ in class count and feature dimension rather than $O(nd)$, which the paper argues fits NISQ-era memory and parameter constraints.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next experiment the paper does not run is to add standard depolarizing or amplitude-damping noise to the SWAP-test circuits to see how much overlap error the logit margin can tolerate; that would directly test whether the NISQ-era framing survives outside simulation.
Because the classifier's logit is linear in the measured overlaps, the model can be viewed as a linear classifier over quantum-state features; this suggests the same prototype scheme could be ported to other distance-based quantum metrics, such as Bures or trace distance, if efficient overlap estimators exist.
The framework's decoupling of a supervised classical head from a frozen quantum backbone is a general recipe: any task with a fixed quantum feature space could use class-conditional mixed-state prototypes, not just image classification.
The rank sweep hints that a per-task or per-class adaptive choice of $K$ could improve average accuracy beyond the fixed $K=7$ used in the main results, at the cost of the paper's deliberately uniform protocol.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The central idea is worth your attention: instead of widening a quantum circuit or reworking the measurement head for every new class, Wu et al. represent each class as a trainable mixed-state prototype and classify by Hilbert-Schmidt distance in a shared 4-qubit density-matrix space. Adding a class then costs one 180-parameter prototype circuit plus K softmax weights, while the 8-qubit backbone never grows. That is a real advance over the usual basis-state readout classifiers, and I don't see it in the cited prior work.

The mathematical core is solid. The decomposition of the HS distance into a purity term and a weighted sum of SWAP-test overlaps (Eqs. 3–8, 12–13) is correct, and the prototype loss indeed reduces to fitting the class-mean density matrix. The softmax parameterization of the mixture weights is a sensible fix for the simplex constraint. Credit also for reporting the rank-K sweep as sensitivity analysis rather than per-dataset tuning, and for explicitly saying the method does not beat strong classical baselines—too many papers hide that.

Now the soft spots, in proportion. The biggest one is that the NISQ claim is untested at its most vulnerable point. The SWAP-test estimates F_i = 2P(0) − 1 are used as logits in Algorithm 1, and the paper's contribution list explicitly claims noise robustness. But every experiment is an exact density-matrix simulation: no noise model, no shot budget, no error mitigation. If depolarizing or readout errors bias P(0), those biases directly shift the logits and can change the argmax. The paper says noise-aware training is future work, which is honest, but then the "noise robustness" in the contributions is not supported. That is a load-bearing gap, not a missing hardware demo.

Second, the incremental tables (Tables II–IV) are single-seed. The static classifier results have four seeds, but the incremental experiments, which are the paper's central claim, do not. Given the variance you see in the four-seed static runs (e.g., Split B ±1.6 points), the incremental differences between ablations could easily be noise. Third, there are no quantum continual learning baselines—no quantum EWC, no quantum replay—so we don't know whether the prototype mechanism is what helps or just the replay.

The complexity claim O(n log^2(max(n,d))) is also underspecified. It appears in Tables II and III without derivation, and it's not obviously the right way to compare quantum prototypes with classical methods. That's a minor issue relative to the noise and seed problems.

Who is this for? Researchers working on quantum machine learning for continual or incremental learning. They will get a coherent framework, a correct derivation, and a clear direction for hardware experiments. It deserves a serious referee. I would send it out with a request for multi-seed incremental runs and at least one noisy simulation or hardware measurement, plus a quantum continual learning baseline. The math is good enough that the idea should not be desk-rejected.

Referee Report

3 major / 6 minor

Summary. This paper proposes a quantum class-incremental learning framework in which a fixed-width 8-qubit QCNN backbone is trained once and each new class is represented by a trainable mixed-state prototype built from the CCPS ansatz on a 4-qubit subsystem. Classification is performed by nearest Hilbert-Schmidt distance between the query density matrix and class prototypes, with the distance decomposed into class-independent purity terms and SWAP-test overlaps. The incremental procedure also maintains a small exemplar buffer of samples closest to the fitted prototypes. Experiments on CIFAR-100 (three 32-class splits) and TinyImageNet compare the method with classical and quantum classifiers and with classical incremental learning baselines, reporting competitive accuracy in both static and incremental settings.

Significance. The analytic core of the paper is sound: Eqs. (3)-(8), (12)-(13), and (27) correctly decompose the Hilbert-Schmidt distance and show that prototype fitting reduces to a variational low-rank approximation of the class-mean density matrix. The proposal to add classes by appending per-class CCPS prototypes rather than widening the quantum circuit directly addresses a real limitation of basis-state-measurement quantum classifiers. The authors also take care to fix the prototype rank K=7 before evaluation and to report the rank sweep as a sensitivity analysis rather than selecting per-dataset maxima. The source code is provided. However, the empirical evidence is currently insufficient to support the paper's NISQ-era and noise-robustness claims, since all experiments are noiseless simulations and the incremental results are single-seed.

major comments (3)

[Section V-A2 / Eq. (6) / Algorithm 1] The paper's central measurement primitive is the SWAP-test overlap F_i = 2P(0)-1 (Eq. (6)), and Algorithm 1 converts these estimates directly into class logits. All experiments, however, are exact density-matrix simulations: Section V-A2 states only that PennyLane circuit simulation is used, with no noise model, shot budget, or error mitigation, and Section VI defers noise-aware training on real devices to future work. Under depolarizing or readout errors, the F_i estimates become biased and noisy, so the logits in Eq. (13) may no longer track the true Hilbert-Schmidt distances, and the argmax in Eq. (9) can change. Since the Abstract and the contribution list in Section I explicitly claim 'noise robustness' and frame the work in the NISQ era, this gap is load-bearing rather than a mere missing hardware demonstration. I recommend adding a noise experiment (e.g., depolarizing noise on the SWAP test with finite shot counts) or substantially revising the claims to noiseless simulation.
[Tables II, III, IV, V] All incremental learning results are reported as single numbers, while only the static classifier comparison (Table I) includes mean±standard deviation over four seeds. The paper's central empirical claim is that the proposed method demonstrates robust representation in incremental learning tasks; without multiple seeds or error bars, we cannot assess the significance of differences such as the 0.5728 Last Acc on Split A compared with, e.g., FeTrIL's 0.6453. Please report mean±std over at least three seeds for Tables II-V, or explicitly label them as single-run results and moderate the robustness claims accordingly.
[Section V-C / Tables II and III] The comparison set for incremental learning contains only classical methods. Given that the introduction cites quantum continual learning results (refs. [33] and [36]), the absence of any quantum continual learning baseline makes it hard to evaluate the contribution relative to prior quantum approaches. Even if those methods are not directly applicable to the same protocol, the authors should explain why they are excluded and, ideally, re-implement or adapt them for comparison; otherwise the claim of a 'feasible direction for quantum incremental learning' is not yet empirically supported.

minor comments (6)

[Abstract] In the sentence 'they have representation capabilities to represent information than a single pure-state prototype', a comparative word appears to be missing; the intended meaning is likely 'richer information than'.
[Section I (Introduction)] The statement 'we append a lightweight, expandable classical module (i.e., an MLP) to manage the incremental class adjustments' is inconsistent with Section IV-B3, which states that the auxiliary MLP head is completely discarded after backbone training; the incremental class adjustments are in fact managed by adding new prototypes. Please rephrase to avoid confusion.
[Section V-A3] The phrase 'By disassembling the modules, a significant amount of storage was saved' is vague; it would be clearer to specify whether the saving is in simulation memory, circuit width, or something else.
[Eq. (27) and Section IV-B4] The text states that optimizing Eq. (27) is equivalent to fitting the class-mean density matrix, but the identity N^(-1)⋅Σ||ρ_n−ρ_c||_F^2 = ||ρ̄−ρ_c||_F^2 + const is not explicitly shown; adding this identity would make the argument easier to verify.
[Tables II and III] The asymptotic notation 'O(nlog 2(max(n,d)))' is ambiguous (base-2 logarithm versus square of the logarithm), and the parameter counting for the prototype circuits (180 parameters per class for a fixed width) should be reconciled with the claimed asymptotic formula.
[Figure 3 caption] The caption 'SWAP test circuit for 2-qubit states ρ and φ' is misleading because Eq. (6) and Algorithm 1 use the SWAP test for general n-qubit states; the caption should be generalized or clarified as a schematic for the single-qubit ancilla case.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the HS-distance logits follow algebraically from the CCPS decomposition, and the reported accuracies are genuine out-of-sample evaluations.

full rationale

The paper's central derivation is self-contained. Equation (7) is obtained from Eq. (3) by substituting Eq. (4), which is the exact purity of a convex mixture of orthogonal basis states, and Eq. (5), which is the linearity of the trace applied to the CCPS expansion; Eq. (6) is the standard SWAP-test measurement identity. Algorithm 1's logit in Eq. (13) is algebraically -1/2 times the squared Hilbert-Schmidt distance up to the query-purity constant Tr[rho(x)^2], so argmax over logits is exactly argmin over D_c(x) in Eq. (9). The prototype loss Eq. (27) is the empirical mean of the same squared HS distance with the fixed query purity removed; thus the training objective and inference rule are consistent but not tautological. The backbone is trained with cross-entropy (Eq. 23) and then frozen; prototypes are fitted to class-conditional density matrices, and test accuracies are evaluated on held-out test sets, so no reported accuracy is the value of a fitted parameter. K=7 is fixed before evaluation (Section V-A2), and the rank sweep in Section V-E is explicitly not used to select per-dataset maxima. The only self-citations are contextual (e.g., [29], [33], [36] in the introduction); none supplies a load-bearing uniqueness theorem or an ansatz that the present derivation needs. The paper's own limitation, noiseless simulation and deferred noise-aware training, undermines the NISQ robustness claim but is a validity gap, not a circularity.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central math borrows CCPS and SWAP tests from prior work and adds no new physical entity. The free parameters are hand-chosen hyperparameters (rank, circuit depth, memory, hidden size, schedule). The axioms are standard quantum information identities plus two domain assumptions the paper does not test: noiseless hardware and sufficient information retention after partial trace.

free parameters (5)

Prototype rank K = 7
Fixed uniformly before evaluation; central to the CCPS prototype equation (Eq 25) and swept in Sec V-E, where the best ranks vary by dataset (12, 7, 15, 7), so the fixed value is a hand-chosen constant that shapes results.
Prototype circuit depth = 3 layers of 4 SU(4) blocks (180 parameters per prototype)
Hardware-efficient tiling chosen in Sec IV-B4; determines expressivity of U_c and the parameter magnitude O(n log).
Exemplar memory budget M = 640
Per-class replay buffer cap used in Algorithm 2; matches FOSTER's 640 while iCaRL and PODNet use 2000, so memory is a chosen experimental constraint.
Auxiliary MLP hidden dimension = 32
Hidden size d_hid in Eq (22), used only during backbone training; affects how well the temporary classifier can shape the density matrix.
Training epochs and learning rates = 240/60/36 epochs, lr 0.01/0.04/0.01
Set in Sec V-A2 for initial task, prototype fitting, and incremental finetuning; standard hyperparameters that influence all reported accuracies.

assumptions (6)

standard math Density matrices admit a spectral decomposition with non-negative weights summing to one (Eq 1).
Background linear algebra used to justify mixed-state representation and the CCPS form.
standard math A unitary U maps an orthonormal basis to an orthonormal basis, so the purity of a CCPS state is the collision probability of the weights, Tr[sigma^2] = sum w_i^2 (Eq 4).
This identity makes the HS distance sampleable as weighted SWAP tests.
standard math The SWAP test gives unbiased overlap estimates in the ideal limit, P(0) = 1/2 + 1/2 Tr[rho sigma] (Eq 6).
Standard quantum measurement identity used in Algorithm 1.
domain assumption Noiseless simulation is a faithful proxy for NISQ hardware behavior.
No noise model, shot noise, or error mitigation is included in Section V-A; the paper invokes NISQ in the abstract and conclusion without testing device noise.
domain assumption The 4-qubit reduced density matrix after QCNN pooling retains enough class-discriminative information to separate 32 classes.
The auxiliary MLP head is discarded, so final accuracy depends on this retention. Empirical results support it, but no proof or information-theoretic guarantee is given.
domain assumption CCPS provides a variational low-rank approximation of the target mixed state under HS distance (from Ezzell et al. [41]).
Borrowed from prior literature; the paper relies on this for the PCA-like noise-filtering interpretation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Quantum Incremental Learning with Mixed State Prototypes." pith.science (2026). https://pith.science/paper/CT2LSIOY

@misc{pith2026260810464,
  author       = {Pith},
  title        = {Pith review of: Quantum Incremental Learning with Mixed State Prototypes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CT2LSIOY}},
  note         = {Machine review of arXiv:2608.10464}
}

read the original abstract

Incremental learning models are required to learn new classes sequentially without catastrophic forgetting, while operating under parameter and memory constraints. In the Noisy Intermediate-Scale Quantum (NISQ) era, although quantum neural networks offer advantages in feature mapping, hardware limitations restrict circuit width. Furthermore, traditional quantum classifiers are constrained by the number of orthogonal basis states, limiting their capacity to accommodate a continually growing number of categories. Thus, we introduce a novel quantum incremental learning framework based on trainable mixed-state prototypes. Its original design incorporates new classes by adding class prototypes rather than increasing the circuit width of the shared quantum backbone. The use of mixed-state prototypes is another key contribution, since they have representation capabilities to represent information than a single pure-state prototype. And the decomposable mixed-state calculation provides lower production costs and a convenient Hilbert-Schmidt (HS) distance metric for classification. Simulation results show that our model achieves high-dimensional feature concentration using a minimal number of qubits, while demonstrating lower computational complexity and robust representation in incremental learning tasks compared with classical baselines.

Figures

Figures reproduced from arXiv: 2608.10464 by the authors.

**Figure 1.** Illustration of incremental learning strategies. (a) Exemplar replay; (b) Regularization; (c) Prototype classification. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** HS distance measurement via CCPS. VQC and SWAP test is shared. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** Overview of the proposed framework. Stage 1: Backbone training. Classical inputs are compressed and amplitude-encoded into a QCNN. The backbone is optimized via an auxiliary head. Stage 2: Prototype Fitting. Basis states are transformed to construct mixed-state prototypes, fitting to the class quantum representations. Stage 3: Memory Update. Samples yielding the minimum HS distance to the prototypes are stored for f… view at source ↗

Figures from the paper (5 more)

**Figure 5.** Figure 5: Overall flowchart of the quantum incremental learning framework. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: The internal gate decomposition of a single two-qubit [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

**Figure 7.** Figure 7: Accuracy at each incremental stage for the prototype, statistics, regularization, and finetuning methods in Table III. Solid, dashed, dash-dotted, and [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

**Figure 8.** Figure 8: Accuracy at each incremental stage for different ablation variants on [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

**Figure 9.** Figure 9: Effect of Prototype Rank K on Avg Acc for CIFAR-100 Split A, Split B, Split C, and TinyImageNet. VI. CONCLUSION In this paper, we introduced a mixed-state quantum prototype framework for class-incremental learning. Within constrained circuit width, our method matches…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 40 canonical work pages

[33]

Quantum continual learning over- coming catastrophic forgetting,

W. Jiang, Z. Lu, and D.-L. Deng, “Quantum continual learning over- coming catastrophic forgetting,”Chinese Physics Letters, vol. 39, no. 5, p. 050303, 2022

work page 2022
[36]

Experimental demonstration of quantum continual learning with superconducting qubits,

C. Zhang, Z. Lu, L. Zhao, S. Xu, W. Li, K. Wang, J. Chen, Y . Wu, F. Jin, X. Zhuet al., “Experimental demonstration of quantum continual learning with superconducting qubits,”npj Quantum Information, 2026

work page 2026
[1]

Class-incremental learning: A survey,

D.-W. Zhou, Q.-W. Wang, Z.-H. Qi, H.-J. Ye, D.-C. Zhan, and Z. Liu, “Class-incremental learning: A survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 9851–9873, 2024

work page 2024
[2]

Few-shot class-incremental learning for classification and object detection: A sur- vey,

J. Zhang, L. Liu, O. Silv ´en, M. Pietik ¨ainen, and D. Hu, “Few-shot class-incremental learning for classification and object detection: A sur- vey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 4, pp. 2924–2945, 2025

work page 2025
[3]

Class-incremental learning: survey and performance evaluation on image classification,

M. Masana, X. Liu, B. Twardowski, M. Menta, A. D. Bagdanov, and J. Van De Weijer, “Class-incremental learning: survey and performance evaluation on image classification,”IEEE Transactions on Pattern Anal- ysis and Machine Intelligence, vol. 45, no. 5, pp. 5513–5533, 2022

work page 2022
[4]

R-dfcil: Relation-guided representation learning for data-free class incremental learning,

Q. Gao, C. Zhao, B. Ghanem, and J. Zhang, “R-dfcil: Relation-guided representation learning for data-free class incremental learning,” in European Conference on Computer Vision. Springer, 2022, pp. 423– 439

work page 2022
[5]

Catastrophic interference in connec- tionist networks: The sequential learning problem,

M. McCloskey and N. J. Cohen, “Catastrophic interference in connec- tionist networks: The sequential learning problem,” inPsychology of learning and motivation. Elsevier, 1989, vol. 24, pp. 109–165

1989
[6]

icarl: Incremental classifier and representation learning,

S.-A. Rebuffi, A. Kolesnikov, G. Sperl, and C. H. Lampert, “icarl: Incremental classifier and representation learning,” inProceedings of the IEEE conference on Computer Vision and Pattern Recognition, 2017, pp. 2001–2010

2017

Show all 59 references

[7]

Foster: Feature boosting and compression for class-incremental learning,

F.-Y . Wang, D.-W. Zhou, H.-J. Ye, and D.-C. Zhan, “Foster: Feature boosting and compression for class-incremental learning,” inEuropean conference on computer vision. Springer, 2022, pp. 398–414

2022
[8]

Podnet: Pooled outputs distillation for small-tasks incremental learning,

A. Douillard, M. Cord, C. Ollion, T. Robert, and E. Valle, “Podnet: Pooled outputs distillation for small-tasks incremental learning,” in European conference on computer vision. Springer, 2020, pp. 86–102

2020
[9]

Overcoming catastrophic forgetting in neural networks,

J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, K. Milan, J. Quan, T. Ramalho, A. Grabska-Barwinska et al., “Overcoming catastrophic forgetting in neural networks,”Pro- ceedings of the national academy of sciences, vol. 114, no. 13, pp. 3521–3526, 2017

2017
[10]

Learning without forgetting,

Z. Li and D. Hoiem, “Learning without forgetting,”IEEE transactions on pattern analysis and machine intelligence, vol. 40, no. 12, pp. 2935– 2947, 2017

2017
[11]

Always be dreaming: A new approach for data-free class-incremental learning,

J. Smith, Y .-C. Hsu, J. Balloch, Y . Shen, H. Jin, and Z. Kira, “Always be dreaming: A new approach for data-free class-incremental learning,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 9374–9384. JOURNAL OF LATEX CLASS FILES, VOL. 14...

2021
[12]

Incomplete data classification via distribution alignment with evidence combination,

L. Huang, J. Fan, S. Wang, G. Liu, and A. W.-C. Liew, “Incomplete data classification via distribution alignment with evidence combination,” Machine Intelligence Research, vol. 23, no. 3, pp. 593–615, 2026

2026
[13]

Dual joint covariance alignment method for incomplete data classification,

L. Huang and G. Liu, “Dual joint covariance alignment method for incomplete data classification,”IEEE Transactions on Neural Networks and Learning Systems, 2026

2026
[14]

Fetril: Feature translation for exemplar-free class-incremental learning,

G. Petit, A. Popescu, H. Schindler, D. Picard, and B. Delezoide, “Fetril: Feature translation for exemplar-free class-incremental learning,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2023, pp. 3911–3920

2023
[15]

Fecam: Exploiting the heterogeneity of class distributions in exemplar-free continual learning,

D. Goswami, Y . Liu, B. Twardowski, and J. Van De Weijer, “Fecam: Exploiting the heterogeneity of class distributions in exemplar-free continual learning,”Advances in Neural Information Processing Systems, vol. 36, pp. 6582–6595, 2023

2023
[16]

Prototype aug- mentation and self-supervision for incremental learning,

F. Zhu, X.-Y . Zhang, C. Wang, F. Yin, and C.-L. Liu, “Prototype aug- mentation and self-supervision for incremental learning,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 5871–5880

2021
[17]

Three types of incremental learning,

G. M. Van de Ven, T. Tuytelaars, and A. S. Tolias, “Three types of incremental learning,”Nature Machine Intelligence, vol. 4, no. 12, pp. 1185–1197, 2022

2022
[18]

Multidomain evolutionary optimization on adversarial link perturbation in imbalanced-size complex systems,

J. Zhao and K. H. Cheong, “Multidomain evolutionary optimization on adversarial link perturbation in imbalanced-size complex systems,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2026

2026
[19]

Quantum classifiers with a trainable kernel,

L. Xu, X.-y. Zhang, M. Li, and S.-q. Shen, “Quantum classifiers with a trainable kernel,”Physical Review Applied, vol. 21, no. 5, p. 054056, 2024

2024
[20]

Ternary coding of maximum deng entropy,

T. Zhan, Y . He, and Y . Deng, “Ternary coding of maximum deng entropy,”Fuzzy Sets and Systems, p. 109913, 2026

2026
[21]

An efficient quan- tum proactive incremental learning algorithm,

L. Li, J. Li, Y . Song, S. Qin, Q. Wen, and F. Gao, “An efficient quan- tum proactive incremental learning algorithm,”Science China Physics, Mechanics & Astronomy, vol. 68, no. 1, p. 210313, 2025

2025
[22]

Quantum option profit and loss encoding for risk assessment,

J. Shi, B. Yuan, Y . Lu, S. Zhang, and X. Li, “Quantum option profit and loss encoding for risk assessment,”IEEE Transactions on Emerging Topics in Computational Intelligence, 2025

2025
[23]

Variational quantum algorithms,

M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincioet al., “Variational quantum algorithms,”Nature Reviews Physics, vol. 3, no. 9, pp. 625– 644, 2021

2021
[24]

An adaptive quantum circuit of dempster’s rule of combination for uncertain pattern classification,

F. Xiao, Y . Zhou, and W. Pedrycz, “An adaptive quantum circuit of dempster’s rule of combination for uncertain pattern classification,”Ad- vances in Neural Information Processing Systems, vol. 38, pp. 17 848– 17 881, 2026

2026
[25]

Model-free quantum gate design and calibration using deep reinforcement learning,

O. Shindi, Q. Yu, P. Girdhar, and D. Dong, “Model-free quantum gate design and calibration using deep reinforcement learning,”IEEE Transactions on Artificial Intelligence, vol. 5, no. 1, pp. 346–357, 2023

2023
[26]

A novel quantum model of mass function for uncertain information fusion,

X. Deng, S. Xue, and W. Jiang, “A novel quantum model of mass function for uncertain information fusion,”Information Fusion, vol. 89, pp. 619–631, 2023

2023
[27]

Negation of the quantum mass function for multisource quantum information fusion with its application to pattern classification,

F. Xiao and W. Pedrycz, “Negation of the quantum mass function for multisource quantum information fusion with its application to pattern classification,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 2, pp. 2054–2070, 2022

2022
[28]

A dependence assessment method based on quantum model of mass function in human reliability analysis,

X. Su, X. Huang, X. Pan, and D. Meng, “A dependence assessment method based on quantum model of mass function in human reliability analysis,”Expert Systems with Applications, vol. 299, p. 129992, 2026

2026
[29]

Feature entanglement- based quantum multimodal fusion neural network,

Y . Wu, Q. Zhou, J. Geng, X. Deng, and W. Jiang, “Feature entanglement- based quantum multimodal fusion neural network,”arXiv preprint arXiv:2601.07856, 2026

2026
[30]

Multimodal remote sensing of thunderstorm charge mo- tion: A radar echo and electric field fusion approach,

X. Yang, H. Xing, F. Zhu, Y . Chen, D. Camacho, X. Dong, and W. Pedrycz, “Multimodal remote sensing of thunderstorm charge mo- tion: A radar echo and electric field fusion approach,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–13, 2025

2025
[31]

Qsyncfold: quantum neural network for multidimensional sync-discovery in protein folding,

J. Shi, P. Du, W. Zeng, W. Wang, S. Peng, and X. Li, “Qsyncfold: quantum neural network for multidimensional sync-discovery in protein folding,”Briefings in Bioinformatics, vol. 27, no. 3, p. bbag234, 2026

2026
[32]

Qsan: A near- term achievable quantum self-attention network,

J. Shi, R.-X. Zhao, W. Wang, S. Zhang, and X. Li, “Qsan: A near- term achievable quantum self-attention network,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 8, pp. 13 995– 14 008, 2024

2024
[34]

Problem-dependent power of quantum neural networks on multiclass classification,

Y . Du, Y . Yang, D. Tao, and M.-H. Hsieh, “Problem-dependent power of quantum neural networks on multiclass classification,”Physical Review Letters, vol. 131, no. 14, p. 140601, 2023

2023
[35]

Unified framework for quantum classification,

N. A. Nghiem, S. Y .-C. Chen, and T.-C. Wei, “Unified framework for quantum classification,”Physical Review Research, vol. 3, no. 3, p. 033056, 2021

2021
[37]

Optimal (controlled) quantum state preparation and improved unitary synthesis by quantum circuits with any number of ancillary qubits,

P. Yuan and S. Zhang, “Optimal (controlled) quantum state preparation and improved unitary synthesis by quantum circuits with any number of ancillary qubits,”Quantum, vol. 7, p. 956, 2023

2023
[38]

Quantum embeddings for machine learning,

S. Lloyd, M. Schuld, A. Ijaz, J. Izaac, and N. Killoran, “Quantum embeddings for machine learning,”arXiv preprint arXiv:2001.03622, 2020

2001 arXiv
[39]

Individual linguis- tic granular computing: A granulation–degranulation-based approach,

T. Huang, Q. Zhang, W. Pedrycz, and S. Yang, “Individual linguis- tic granular computing: A granulation–degranulation-based approach,” IEEE Transactions on Cybernetics, 2026

2026
[40]

Learning- based quantum robust control: algorithm, applications, and experiments,

D. Dong, X. Xing, H. Ma, C. Chen, Z. Liu, and H. Rabitz, “Learning- based quantum robust control: algorithm, applications, and experiments,” IEEE transactions on cybernetics, vol. 50, no. 8, pp. 3581–3593, 2019

2019
[41]

Quantum mixed state compiling,

N. Ezzell, E. M. Ball, A. U. Siddiqui, M. M. Wilde, A. T. Sornborger, P. J. Coles, and Z. Holmes, “Quantum mixed state compiling,”Quantum Science and Technology, vol. 8, no. 3, p. 035001, 2023

2023
[42]

Experimental measurement of the hilbert-schmidt distance between two-qubit states as a means for reducing the complexity of machine learning,

V . Tr´avn´ıˇcek, K. Bartkiewicz, A. ˇCernoch, and K. Lemr, “Experimental measurement of the hilbert-schmidt distance between two-qubit states as a means for reducing the complexity of machine learning,”Physical review letters, vol. 123, no. 26, p. 260501, 2019

2019
[43]

Prototype-sample relation distillation: towards replay-free continual learning,

N. Asadi, M. Davari, S. Mudur, R. Aljundi, and E. Belilovsky, “Prototype-sample relation distillation: towards replay-free continual learning,” inInternational conference on machine learning. PMLR, 2023, pp. 1093–1106

2023
[44]

Variational quantum fidelity estimation,

M. Cerezo, A. Poremba, L. Cincio, and P. J. Coles, “Variational quantum fidelity estimation,”Quantum, vol. 4, p. 248, 2020

2020
[45]

Quantum convolutional neural networks,

I. Cong, S. Choi, and M. D. Lukin, “Quantum convolutional neural networks,”Nature Physics, vol. 15, no. 12, pp. 1273–1278, 2019

2019
[46]

Quantum circuit architecture search for variational quantum algorithms,

Y . Du, T. Huang, S. You, M.-H. Hsieh, and D. Tao, “Quantum circuit architecture search for variational quantum algorithms,”npj Quantum Information, vol. 8, no. 1, p. 62, 2022

2022
[47]

Branching quantum convolutional neural networks,

I. MacCormack, C. Delaney, A. Galda, N. Aggarwal, and P. Narang, “Branching quantum convolutional neural networks,”Physical Review Research, vol. 4, no. 1, p. 013117, 2022

2022
[48]

Circuit-centric quantum classifiers,

M. Schuld, A. Bocharov, K. M. Svore, and N. Wiebe, “Circuit-centric quantum classifiers,”Physical Review A, vol. 101, no. 3, p. 032308, 2020

2020
[49]

Hierarchical quantum classifiers,

E. Grant, M. Benedetti, S. Cao, A. Hallam, J. Lockhart, V . Stojevic, A. G. Green, and S. Severini, “Hierarchical quantum classifiers,”npj Quantum Information, vol. 4, no. 1, p. 65, 2018

2018
[50]

Quclassi: A hybrid deep neural network architecture based on quantum state fidelity,

S. A. Stein, B. Baheri, D. Chen, Y . Mao, Q. Guan, A. Li, S. Xu, and C. Ding, “Quclassi: A hybrid deep neural network architecture based on quantum state fidelity,”Proceedings of Machine Learning and Systems, vol. 4, pp. 251–264, 2022

2022
[51]

Large scale incremental learning,

Y . Wu, Y . Chen, L. Wang, Y . Ye, Z. Liu, Y . Guo, and Y . Fu, “Large scale incremental learning,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 374–382

2019
[52]

Self-sustaining representation expansion for non-exemplar class-incremental learning,

K. Zhu, W. Zhai, Y . Cao, J. Luo, and Z.-J. Zha, “Self-sustaining representation expansion for non-exemplar class-incremental learning,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 9296–9305

2022
[53]

Dark experience for general continual learning: a strong, simple baseline,

P. Buzzega, M. Boschini, A. Porrello, D. Abati, and S. Calderara, “Dark experience for general continual learning: a strong, simple baseline,” Advances in neural information processing systems, vol. 33, pp. 15 920– 15 930, 2020

2020
[54]

Learning a unified classifier incrementally via rebalancing,

S. Hou, X. Pan, C. C. Loy, Z. Wang, and D. Lin, “Learning a unified classifier incrementally via rebalancing,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 831–839

2019
[55]

Metric learning for large scale image classification: Generalizing to new classes at near- zero cost,

T. Mensink, J. Verbeek, F. Perronnin, and G. Csurka, “Metric learning for large scale image classification: Generalizing to new classes at near- zero cost,” inEuropean conference on computer vision. Springer, 2012, pp. 488–501

2012
[56]

Few-shot class-incremental learning,

X. Tao, X. Hong, X. Chang, S. Dong, X. Wei, and Y . Gong, “Few-shot class-incremental learning,” in2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020, pp. 12 180– 12 189

2020
[57]

Semantic drift compensation for class-incremental learning,

L. Yu, B. Twardowski, X. Liu, L. Herranz, K. Wang, Y . Cheng, S. Jui, and J. Van de Weijer, “Semantic drift compensation for class-incremental learning,” in2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020, pp. 6980–6989. JOURNAL OF LATEX C...

2020
[58]

Deesil: Deep-shallow incremental learn- ing

E. Belouadah and A. Popescu, “Deesil: Deep-shallow incremental learn- ing.” inProceedings of the European conference on computer vision (ECCV) workshops, 2018, pp. 0–0

2018
[59]

More classifiers, less forgetting: A generic multi-classifier paradigm for incremental learning,

Y . Liu, S. Parisot, G. Slabaugh, X. Jia, A. Leonardis, and T. Tuytelaars, “More classifiers, less forgetting: A generic multi-classifier paradigm for incremental learning,” inEuropean Conference on Computer Vision. Springer, 2020, pp. 699–716

2020

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [33]

Quantum continual learning over- coming catastrophic forgetting,

W. Jiang, Z. Lu, and D.-L. Deng, “Quantum continual learning over- coming catastrophic forgetting,”Chinese Physics Letters, vol. 39, no. 5, p. 050303, 2022

work page 2022

[2] [36]

Experimental demonstration of quantum continual learning with superconducting qubits,

C. Zhang, Z. Lu, L. Zhao, S. Xu, W. Li, K. Wang, J. Chen, Y . Wu, F. Jin, X. Zhuet al., “Experimental demonstration of quantum continual learning with superconducting qubits,”npj Quantum Information, 2026

work page 2026

[3] [1]

Class-incremental learning: A survey,

D.-W. Zhou, Q.-W. Wang, Z.-H. Qi, H.-J. Ye, D.-C. Zhan, and Z. Liu, “Class-incremental learning: A survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 9851–9873, 2024

work page 2024

[4] [2]

Few-shot class-incremental learning for classification and object detection: A sur- vey,

J. Zhang, L. Liu, O. Silv ´en, M. Pietik ¨ainen, and D. Hu, “Few-shot class-incremental learning for classification and object detection: A sur- vey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 4, pp. 2924–2945, 2025

work page 2025

[5] [3]

Class-incremental learning: survey and performance evaluation on image classification,

M. Masana, X. Liu, B. Twardowski, M. Menta, A. D. Bagdanov, and J. Van De Weijer, “Class-incremental learning: survey and performance evaluation on image classification,”IEEE Transactions on Pattern Anal- ysis and Machine Intelligence, vol. 45, no. 5, pp. 5513–5533, 2022

work page 2022

[6] [4]

R-dfcil: Relation-guided representation learning for data-free class incremental learning,

Q. Gao, C. Zhao, B. Ghanem, and J. Zhang, “R-dfcil: Relation-guided representation learning for data-free class incremental learning,” in European Conference on Computer Vision. Springer, 2022, pp. 423– 439

work page 2022

[7] [5]

Catastrophic interference in connec- tionist networks: The sequential learning problem,

M. McCloskey and N. J. Cohen, “Catastrophic interference in connec- tionist networks: The sequential learning problem,” inPsychology of learning and motivation. Elsevier, 1989, vol. 24, pp. 109–165

1989

[8] [6]

icarl: Incremental classifier and representation learning,

S.-A. Rebuffi, A. Kolesnikov, G. Sperl, and C. H. Lampert, “icarl: Incremental classifier and representation learning,” inProceedings of the IEEE conference on Computer Vision and Pattern Recognition, 2017, pp. 2001–2010

2017