Constraint-aware Learning of Probabilistic Sequential Models for Multi-Label Classification

Anna Polova; Michael Benedikt; Mykhailo Buleshnyi; Zsolt Zombori

REVIEW 4 major objections 5 minor 20 references

Constraint-aware Learning of Probabilistic Sequential Models for Multi-Label Classification

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A two-stage model with per-label classifiers feeding a sequential integrator outperforms a constraint-encoding baseline on 8 of 9 datasets and learns logical constraints from data alone.

desk verdict A promising new two-stage architecture for constraint-aware multi-label classification, but the headline comparison confounds architecture with capacity and hyperparameters, so the central architectural claim still needs a cleaner control. read the letter →

arxiv 2507.15156 v1 pith:ENX4YK6A submitted 2025-07-20 cs.LG cs.AIcs.LO

Mykhailo Buleshnyi , Anna Polova , Zsolt Zombori , Michael Benedikt This is my paper · ORCID

classification cs.LGcs.AIcs.LO

keywords multi-labelclassificationlogicalconstraintssequentialmodelbeamsearchsemi-supervisedlearningSAT-basedinferencelabelcorrelationnext-tokenprediction

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

This paper asks whether multi-label classification over many labels can be improved by replacing the usual single-pass output layer with a two-stage architecture: a base network predicts a marginal probability for each label, and a sequential integrator reads those marginals one label at a time to produce a joint distribution over all label combinations. The authors' central claim is that this structure lets the model capture correlations between labels, including correlations that correspond to logical constraints on the output space, without being given the constraints explicitly. They report that the BaseSeq model outperforms the base model and an explicit constraint-encoding baseline (CCN) on 8 of the 9 informative datasets, often by a large margin, and that a SAT-solver-based variant enforces constraints with almost no accuracy change. They also show that beam search approximates full-distribution inference well, and that exploiting constraints on unlabeled data is only reliably helpful when a solid supervised foundation already exists.

What carries the argument

The central object is the prefix conditional model \(C\), a network that takes a vector of per-label marginal probabilities together with a partial assignment to the first \(j\) labels and returns the probability that label \(j+1\) is true. Multiplying these conditional probabilities over positions defines a full joint distribution over label valuations, giving the sequential model \(\text{Seq}_C\); combined with a base model \(B\), this is the BaseSeq architecture. Inference is done by beam search over prefixes, and constraint enforcement is done by checking each beam's prefix for satisfiability with a SAT solver and pruning unsatisfiable prefixes. For semi-supervised training, the paper introduces a constraint loss that penalizes only the suffix decisions of sampled invalid valuations that are not shared with any sampled valid valuation.

What would settle it

Train parameter-matched versions of the Base and CCN models with hidden widths inflated to reach the same total parameter count as BaseSeq on Emotions, Yeast, and Arts; if those baselines close the accuracy gap, the paper's claim that the sequential architecture itself drives the improvement would be refuted.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes a new empirical role for sequential integrators in multi-label classification: the BaseSeq model — a per-label base network followed by a prefix-conditional sequential model — yields higher exact-match and top-k accuracy than both the independence-style base model and the constraint-specific CCN layer on nearly every dataset considered. The same experiments show that the sequential model internalizes constraints: on datasets where the base model violates constraints in 10–20% of predictions, BaseSeq drops the violation rate to zero or near zero while keeping accuracy, and the SAT-based BaseSeqS version changes accuracy barely at all while enforcing all constraints. The authors conclude that such models can learn complex constraints from data, and that, perhaps as a consequence, there is no consistent significant advantage to adding constraint-based training signals.

Load-bearing premise

The main comparisons keep the sequential integrator at 300 hidden units while the Base and CCN baselines are much smaller networks, so the reported accuracy gains could in principle come from the integrator's extra parameters rather than from its sequential structure; the ablation in Section 4.7 changes only the base model's depth, not the integrator's capacity.

Editorial extensions

If this is right

On nine datasets where the base model is not already perfect, switching the integrator from a constraint-specific layer to a sequential model improves exact-match accuracy on eight, with gains of several percentage points over both Base and CCN on datasets such as Arts, Emotions, and Yeast.
The sequential model learns many logical constraints from data: on Emotions, Yeast, Scene, and Image, where the base model violates constraints in 10–20% of predictions, BaseSeq's violation rate falls to zero or near zero without any constraint input.
Approximate inference is sufficient in practice: beam search with width 4 matches the top-k accuracy of exact full-distribution inference on Yeast, Scene, and Emotions, and performance plateaus by width 4, so the model need not pay the exponential cost of exact inference.
Semi-supervised exploitation of constraints is not uniformly beneficial: pseudo-labeling is stable and conservative, while constraint loss can produce large gains at moderate supervision (5% on Medical at 50% supervision) but can also collapse accuracy at low supervision (70% drop on Genbase).
A SAT-solver filter can turn a model that already satisfies constraints most of the time into one that always satisfies them, at no measurable accuracy cost on the benchmarks.

Reading between the lines

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

If the sequential integrator genuinely internalizes constraints, then in very large label spaces where explicit constraint extraction is impractical, a trained BaseSeq could act as an implicit constraint learner with the SAT filter as a safety net rather than as the primary mechanism.
The reported sensitivity to label ordering on the Medical dataset suggests a concrete next experiment: train with random or learned label orderings and check whether constraint learning and accuracy become order-independent; the paper leaves this open.
The constraint-loss recipe can be operationalized as a monitoring rule: track the fraction of beams that satisfy constraints during training and the overlap between valid and invalid beam prefixes; this would directly test whether the overlap is what makes constraint loss help or hurt.

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

Quick take: this is a genuinely new architecture for constraint-aware multi-label classification, and the authors are honest that the results are preliminary. The soft spot is the classic one: the headline comparison doesn't separate architecture from capacity and tuning.

What's new: a two-stage model where a per-label network feeds marginals into an autoregressive prefix-conditional integrator, trained with NLL, with beam search for inference, SAT-based pruning for hard constraint enforcement, and two constraint-aware unsupervised losses. The pieces are known, but the combination is new, and the empirical comparison on the 11 datasets from [4] is a fair test on standard benchmarks. The paper is well-written and doesn't oversell: it explicitly calls the experiments preliminary, notes the datasets are small with simple constraints, and reports negative results in the unsupervised setting (e.g., the Genbase 70% drop). The artifact is available, which helps reproducibility.

The main weakness is exactly what the stress-test note says: Table 1 confounds the architecture with extra parameters and a different hyperparameter regime. BaseSeq adds a 300-unit two-layer integrator, and dropout, learning rate, and batch size all differ from the Base/CCN configuration. Section 4.7 varies base depth but keeps the integrator fixed, so it doesn't address the capacity confound. A prefix-blind integrator control (same integrator, but conditioned only on the current marginal, not the prefix) would be the cleanest test of whether the sequential mechanism is what matters. The toy experiment in Appendix A is matched-capacity and shows a small advantage for seq-only over CCN, but it has only two output variables and the seq-only model fails on real datasets, so it's not enough. The constraint-learning claim in Section 5 is also stronger than the evidence: on Medical, where constraints are simple implications, CCN beats BaseSeq, so the architecture's advantage isn't universal.

That said, the paper is honest about the preliminary nature and the comparison gap. The core idea is promising and the empirical pattern (near-zero constraint violation without explicit constraints, beam search plateaus at width 4) is worth taking seriously. More rigorous evaluation with capacity controls and error bars would make this a solid paper.

Recommendation: send it to peer review. The architecture is new and the question—can a sequential model learn label correlations that look like logical constraints—is worth getting right. I'd bring it to reading group.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a two-stage architecture for multi-label classification: a base network predicts per-label marginal probabilities, and a sequential integrator (a prefix-conditional model) converts these marginals into a joint distribution over output valuations. Inference is approximated by beam search, constraint satisfaction can be guaranteed by integrating a SAT solver into beam search, and unsupervised data can be exploited through pseudo-labeling or a constraint-loss objective. Experiments on the 11 datasets of Giunchiglia and Lukasiewicz compare the proposed BaseSeq model with the Base model and the constraint-encoding CCN baseline in terms of exact-match accuracy, target probability, top-k accuracy, and constraint violation. The paper claims that the sequential architecture learns label correlations and constraints from data, often outperforming CCN without explicit constraint encoding, and that SAT-based pruning provides guaranteed constraint satisfaction with little accuracy change.

Significance. If the central empirical claims are supported, the proposed two-stage design is an attractive and simple way to capture label dependencies without hand-encoding constraints, and the SAT-based enforcement mechanism is a practical contribution. The paper provides code, a capacity-matched toy comparison in Appendix A, beam-width ablations, and an ablation over base-model depth, which are useful elements. However, the headline interpretation that the gains come from the sequential architecture rather than from additional parameters or different training settings is currently under-supported, and most tables lack uncertainty estimates. With the additional controls described below, the paper could make a solid case for the architectural advantage.

major comments (4)

[Section 4, 'Training Hyperparameters' and Table 1] The comparison between BaseSeq and the baselines confounds architecture with training configuration. Base and CCN use learning rate 1e-4, weight decay 1e-4, dropout 0.8, and batch size 4, while BaseSeq uses learning rate 1e-3, weight decay 1e-3, dropout 0.1, and batch size 16. The base model inside BaseSeq is therefore not trained under the same optimization conditions as the standalone Base and CCN models. Before the accuracy margins in Table 1 can be attributed to the sequential integrator, the authors should report results with all models trained under identical hyperparameter settings, or perform a hyperparameter search for each model and report the best configuration. A large dropout difference (0.8 vs. 0.1) alone can substantially change effective model capacity and regularization.
[Section 4.7 and Appendix C] The claim in Section 4.7 that the improvement 'is not due to its extra parameters, but rather to the better architecture' is not supported by the experiment. Varying only the number of base-model layers while keeping the 300-unit integrator fixed does not construct a non-sequential model with comparable total parameter count or comparable training configuration. The integrator alone adds on the order of 10^5 parameters for the output sizes considered, while Base adds none and CCN's layer is non-parametric. A proper control would be, for example, a prefix-blind integrator of the same size and training configuration that receives the same marginal probabilities but is order-invariant, or a non-sequential MLP head of the same capacity added to Base. The Appendix A toy experiment is capacity-matched but uses only two output variables and the Seq-only variant is reported to fail on the real datasets, so it does not resolve the confounding on the main benchmarks.
[Tables 1, 2, 5, 6, and 7] The main empirical tables report point estimates without error bars or confidence intervals. Table 1 reports a single number per model and dataset, while Section 4.7 reports averages of three runs without variance. Several reported margins are small enough that they could be within run-to-run noise (for example, Medical: CCN 0.538 vs. BaseSeq 0.515; Business: BaseSeq 0.589 vs. CCN 0.583; Genbase: BaseSeq 0.985 vs. CCN 0.988). The claim in Section 4.1 that BaseSeq 'outperforms both the Base model and CCN on 8 of the 9 datasets' needs repeated-seed results with standard deviations and, ideally, a paired significance test to be convincing.
[Section 4.4] The default beam width of 4 is justified by the beam-width study in Figure 1, which evaluates accuracy on the same datasets used for the main results. Selecting an inference hyperparameter on the test sets can inflate the reported numbers for BaseSeq relative to methods that do not have this tuning step. The beam width should be selected on a validation split, or the sensitivity analysis should be presented as a selection procedure with the chosen value fixed before evaluating on the test sets.

minor comments (5)

[Abstract] The phrase 'its ability to modelling correlations' should read 'its ability to model correlations'.
[Example 3.1] The second term in the displayed loss should be log(1 - C(pa, (1))), not (1 - log(C(pa, (1)))). The preceding formula is correct, but the example as written is inconsistent with it.
[References, [12]] Reference [12] is cited for PySAT, but the reference points to the Python satellite data analysis toolkit 'pysat' by Stoneback et al. The correct citation for the PySAT SAT solver should be used.
[Section 3.3] The pseudo-labeling procedure is a self-training loop in which the model's own beam search generates the targets. This is standard pseudo-labeling, but the phrase 'unsupervised learning' should be qualified to avoid implying that the targets are independent of the model being trained.
[Section 4.1] The discussion of the Medical dataset mentions that 'preliminary experiments suggest' variable ordering changes accuracy by 1-2%, but no data or protocol for those experiments is provided. Reporting these results in a table or appendix would make the claim checkable.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's central claims are empirical comparisons against external ground truth and held-out constraints; its self-training components are standard semi-supervised techniques, not circularly validated derivations.

full rationale

The paper's central claims are empirical: BaseSeq outperforms baselines on external test sets (Table 1) and learns constraints, evidenced by low constraint-violation rates on held-out data without explicit constraint supervision (Table 5). These claims are checked against independent ground-truth labels and constraints, not against the model's own training objectives. The architecture is defined by a standard product-of-conditionals factorization, and the supervised loss is the corresponding negative log-likelihood; no equation reduces to its own input. The pseudo-labeling method in Section 3.3 is a self-training loop (the model's beam search generates the pseudo labels), but the paper explicitly attributes this technique to prior work [8] and evaluates it by external accuracy, not by consistency with its own outputs. The constraint loss also uses the model's own beam-search samples as a training signal, but this is a training procedure, not a claimed derivation or prediction. The SAT-based enforcement in BaseSeqS guarantees constraint satisfaction by construction, which the paper states explicitly; it does not present this guarantee as a learned capability. The inference that BaseSeq internally learns constraints is supported by the external constraint-violation metric on test data, not by construction. The only self-citations in the paper (e.g., reference [20]) appear in related-work discussion and are not load-bearing for the claims. The skeptic's concern about parameter-count confounding between BaseSeq and baselines is a validity threat to the empirical comparison, not a circularity: it does not make any prediction equivalent to its inputs by definition. Overall, no circular step can be exhibited: no fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors, and no ansatz is smuggled in via citation. Score 0.

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

No new physical or formal entities are introduced; the sequential integrator is an architectural artifact. The central empirical comparison rests on a small set of hyperparameters and domain assumptions about the datasets, the propositional constraint format, and the reliability of beam search.

free parameters (5)

Beam width (evaluation) = 4 by default
Selected from the beam-width sensitivity experiment in Section 4.4 on the same datasets; all main results use this value, and the paper reports that accuracy plateaus near k=4.
Beam width (unsupervised training) = 5
Used in Section 4.6 for pseudo-labeling and constraint loss; the paper does not report a dedicated tuning study for this setting.
Sequential integrator size = 2 layers, 300 hidden units
Fixed in Section 4 for BaseSeq and BaseSeqS with no sensitivity analysis; this large capacity may partly explain the accuracy advantage over the smaller Base and CCN models.
Sequential model training hyperparameters = lr=0.001, weight_decay=0.001, dropout=0.1, batch_size=16
Default configuration for the sequential model in Section 4; it differs from the Base/CCN settings, which are taken from prior work.
Base and CCN training hyperparameters = lr=0.0001, weight_decay=0.0001, dropout=0.8, batch_size=4
Borrowed from [4] and not re-tuned here, so the cross-model comparison depends on the prior paper's configuration.

assumptions (5)

standard math The joint distribution over output valuations is the product of prefix conditional probabilities, as in Definition 3.
This is the probability chain rule and correctly underlies the sequential model, but it assumes the network outputs are interpretable as conditional probabilities.
domain assumption Constraints are input-independent propositional formulas over output labels.
Used throughout Sections 2, 3 and 4 for pseudo-labeling, constraint loss, and SAT-based pruning; sample-dependent or hidden-label constraints are explicitly out of scope.
domain assumption Beam search with small width yields high-probability valuations close to exact inference.
This is load-bearing for approximate inference and for generating pseudo-labels in Section 3.3; Section 4.3 checks it only for three datasets with small output spaces.
domain assumption The 11 datasets from [4] are representative for evaluating constrained multi-label classification.
All experiments use these tabular datasets; the conclusion itself cautions that the constraints are simple and the data are limited, so generality is untested.
domain assumption Pseudo-labels produced by the model's own beam search are reliable enough to use as training targets.
Section 3.3 pseudo-labeling depends on this premise, and the results in Tables 6 and 7 show that the unsupervised methods are often unstable or harmful at low supervision.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Constraint-aware Learning of Probabilistic Sequential Models for Multi-Label Classification." pith.science (2026). https://pith.science/paper/ENX4YK6A

@misc{pith2026250715156,
  author       = {Pith},
  title        = {Pith review of: Constraint-aware Learning of Probabilistic Sequential Models for Multi-Label Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ENX4YK6A}},
  note         = {Machine review of arXiv:2507.15156}
}

read the original abstract

We investigate multi-label classification involving large sets of labels, where the output labels may be known to satisfy some logical constraints. We look at an architecture in which classifiers for individual labels are fed into an expressive sequential model, which produces a joint distribution. One of the potential advantages for such an expressive model is its ability to modelling correlations, as can arise from constraints. We empirically demonstrate the ability of the architecture both to exploit constraints in training and to enforce constraints at inference time.

Figures

Figures reproduced from arXiv: 2507.15156 by the authors.

**Figure 2.** Comparing a Seq-only (left) model with CCN (right) on the small syntactic problem. In each subplot, the first column corresponds to 𝑂1 and the second to 𝑂2. Green and yellow rectangles show the grand truth. For each input, color indicates the models probability estimate for the given class. Rows represent different target variable scenarios: complete overlap, partial overlap, and disjoint [PITH_FULL_IMAGE:figures/f… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 19 canonical work pages

[4]

Giunchiglia, E., Lukasiewicz, T.: Multi-label classification neural networks with hard logical constraints. J. Artif. Int. Res. 72, 759–818 (Jan 2022)

work page 2022
[1]

Camillo Maria Carusoa and Paolo Sodaa, and Valerio Guarrasia: Not another imputation method: A transformer-based model for missing values in tabular datasets (2024), https://arxiv.org/pdf/2407.11540

arXiv 2024
[2]

In: ICML (2019)

Collobert, R., Hannun, A., Synnaeve, G.: A fully differentiable beam search decoder. In: ICML (2019)

work page 2019
[3]

In: International Conference on Machine Learning (2019)

Frogner, C., Poggio, T.: Fast and flexible inference of joint distributions from their marginals. In: International Conference on Machine Learning (2019)

work page 2019
[5]

Giunchiglia, E., Stoian, M.C., Khan, S., Cuzzolin, F., Lukasiewicz, T.: ROAD-R: the autonomous driving dataset with logical requirements. Mach. Learn. 112(9), 3261–3291 (2023)

work page 2023
[6]

In: IJCAI (2022)

Giunchiglia, E., Stoian, M.C., Lukasiewicz, T.: Deep learning with logical con- straints. In: IJCAI (2022)

work page 2022
[7]

In: ICML (2010)

Hariharan, B., Zelnik-Manor, L., Vishwanathan, S.V.N., Varma, M.: Large scale max-margin multi-label classification with priors. In: ICML (2010)

work page 2010
[8]

In: Workshop on challenges in representation learning (2013)

Lee, D.H.: Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In: Workshop on challenges in representation learning (2013)

work page 2013

Show all 20 references

[9]

In: ICML (2024)

Liu, A., Ahmed, K., den Broeck, G.V.: Scaling tractable probabilistic circuits: A systems perspective. In: ICML (2024)

2024
[10]

Artificial Intelligence 298, 103504 (2021)

Manhaeve, R., Dumančić, S., Kimmig, A., Demeester, T., De Raedt, L.: Neural probabilistic logic programming in DeepProbLog. Artificial Intelligence 298, 103504 (2021)

2021
[11]

International Journal of Production Research 26(1), 35–62 (1988)

OW, P.S., MORTON, T.E.: Filtered beam search in scheduling. International Journal of Production Research 26(1), 35–62 (1988)

1988
[12]

https://doi.org/10.5281/zenodo.1199703, https://doi.org/10.5281/zenodo.1199703

Stoneback, R., Klenzing, J., Burrell, A., Spence, C., Depew, M., Hargrave, N., Smith, J., von Bose, V., Pembroke, A., Iyer, G., Luis, S.: Python satellite data analysis toolkit (pysat) vx.y.z (2021). https://doi.org/10.5281/zenodo.1199703, https://doi.org/10.5281/zenodo.1199703

2021 doi
[13]

Trauger, J., Tewari, A.: On Next-Token Prediction in LLMs: How End Goals Determine the Consistency of Decoding Algorithms (2025), https://arxiv.org/ abs/2505.11183

2025 arXiv
[14]

IEEE Transactions on Audio, Speech, and Language Processing 16(2) (2008)

Turnbull, D., Barrington, L., Torres, D., Lanckriet, G.: Semantic annotation and retrieval of music and sound effects. IEEE Transactions on Audio, Speech, and Language Processing 16(2) (2008)

2008
[15]

In: NeurIPS (2024)

Wang, B., Mauá, D.D., den Broeck, G.V., Choi, Y.: A compositional atlas for algebraic circuits. In: NeurIPS (2024)

2024
[16]

In: ICML (2023)

Wang, K., He, H., Nguyen, T.D., Kumar, P., Roth, D.: On regularization and inference with label constraints. In: ICML (2023)

2023
[17]

In: NeurIPS (2023)

Wang, K., Tsamoura, E., Roth, D.: On learning latent models with multi-instance weak supervision. In: NeurIPS (2023)

2023
[18]

CoRR abs/2407.10000 (2024)

Wang, K., Tsamoura, E., Roth, D.: On characterizing and mitigating imbalances in multi-instance partial label learning. CoRR abs/2407.10000 (2024)

2024
[19]

In: ICML (2018)

Xu, J., Zhang, Z., Friedman, T., Liang, Y., Van den Broeck, G.: A semantic loss function for deep learning with symbolic knowledge. In: ICML (2018)

2018
[20]

JMLR 25(1) (2024) 8 A THE IMPACT OF THE SEQUENTIAL ARCHITECTURE ON A TOY EXAMPLE We illustrate the power of the sequential architecture with respect to CCN on a small problem

Zombori, Z., Rissaki, A., Szabó, K., Gatterbauer, W., Benedikt, M.: Towards unbi- ased exploration in partial label learning. JMLR 25(1) (2024) 8 A THE IMPACT OF THE SEQUENTIAL ARCHITECTURE ON A TOY EXAMPLE We illustrate the power of the sequential architecture with respect to...

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [4]

Giunchiglia, E., Lukasiewicz, T.: Multi-label classification neural networks with hard logical constraints. J. Artif. Int. Res. 72, 759–818 (Jan 2022)

work page 2022

[2] [1]

Camillo Maria Carusoa and Paolo Sodaa, and Valerio Guarrasia: Not another imputation method: A transformer-based model for missing values in tabular datasets (2024), https://arxiv.org/pdf/2407.11540

arXiv 2024

[3] [2]

In: ICML (2019)

Collobert, R., Hannun, A., Synnaeve, G.: A fully differentiable beam search decoder. In: ICML (2019)

work page 2019

[4] [3]

In: International Conference on Machine Learning (2019)

Frogner, C., Poggio, T.: Fast and flexible inference of joint distributions from their marginals. In: International Conference on Machine Learning (2019)

work page 2019

[5] [5]

Giunchiglia, E., Stoian, M.C., Khan, S., Cuzzolin, F., Lukasiewicz, T.: ROAD-R: the autonomous driving dataset with logical requirements. Mach. Learn. 112(9), 3261–3291 (2023)

work page 2023

[6] [6]

In: IJCAI (2022)

Giunchiglia, E., Stoian, M.C., Lukasiewicz, T.: Deep learning with logical con- straints. In: IJCAI (2022)

work page 2022

[7] [7]

In: ICML (2010)

Hariharan, B., Zelnik-Manor, L., Vishwanathan, S.V.N., Varma, M.: Large scale max-margin multi-label classification with priors. In: ICML (2010)

work page 2010

[8] [8]

In: Workshop on challenges in representation learning (2013)

Lee, D.H.: Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In: Workshop on challenges in representation learning (2013)

work page 2013