Pith. sign in

REVIEW 3 major objections 4 minor 77 references

One Knob to Rule Them All: A Unified Optimal Transport View of Cold-Start Active Learning

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that typicality-, coverage-, and diversity-based cold-start selectors are one transport program in disguise, and that one data-tuned temperature, $\varepsilon^* = c\hat{m}(b)$, makes the Sinkhorn selector minimax-rate-optim

desk verdict A genuinely useful unification of cold-start active learning with a clean minimax bound, but the universal constant c is an unverified load-bearing assumption and the empirical dominance rests on thin seeds. read the letter →

arxiv 2608.03249 v1 pith:EQZTJM4Y submitted 2026-08-04 cs.AI

Ning Zhu , Xiaochuan Ma , Juntao Xu , Jingze Liang , Mengfei Zhao , An Chen , Liang-Jian Deng This is my paper · ORCID
classification cs.AI
keywords cold-startactivelearningoptimaltransportSinkhornscalingentropicregularizationminimaxrateclosest-anchorcostintrinsicdimensiontask-agnosticselection
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

Cold-start active learning must pick the first annotation batch with no labels and no trained model, forcing each existing method to commit to a fixed geometric bias — typicality, coverage, or diversity — that wins on some datasets and fails on others. The paper's claim is that those three routes are all the same allocation program viewed through different lenses, and that the only quantity needing adaptation is the entropic regularization strength $\varepsilon$ of a balanced Sinkhorn coupling. The theory pins the right scale: $\varepsilon$ should equal a dimensionless constant times the empirical mean closest-anchor cost, the unique scale-equivariant statistic, and with this choice the selector attains the minimax rate $\Theta(b^{-2/d_{\mathrm{int}}})$ — optimal label efficiency in the budget $b$ given the intrinsic dimension $d_{\mathrm{int}}$ of the feature manifold. Empirically the one rule, $\varepsilon$-AS, ranks first in 25 of 26 dataset-budget settings across six datasets and beats the strongest fixed-rule baseline on ImageNet-1k by 1.29% average accuracy while selecting 2.3$\times$ faster. If true, cold-start selection needs no dataset-specific heuristic: unlabeled geometry alone sets the allocation scale.

What carries the argument

The central object is the balanced entropic transport program $J^R_\varepsilon(\theta,\pi) = \langle\pi, C_\theta\rangle + \varepsilon\,\mathrm{KL}(\pi \| a\otimes u) + \Omega^R(\theta)$, together with a realization $R = (\Theta^R, \Pi^R, C^R, \Omega^R, D^R)$ fixing the allowed representatives, allocation constraints, cost, support energy, and decoder. Theorem 1 shows this one program exactly subsumes TypiClust, ProbCover, and ActiveFT, converting a choice among methods into a choice of one scalar $\varepsilon$. The carrying identity is the adaptive temperature $\varepsilon^* = c\,\hat{m}(b)$, where $\hat{m}(b)$ is the empirical mean closest-anchor cost; cost-scale equivariance of Sinkhorn p

What would settle it

Re-run the calibration grid ($c \in \{0.1, 0.2, \dots, 0.9\}$) on labels from a target dataset such as CIFAR-100 or ImageNet-1k: if the best-performing $c$ is far from $0.4$ there, the claim that one constant transfers across pool geometries fails. A direct check of the decoder-fidelity assumption is also possible: compute the probability-weighted anchor-to-query displacement $\sum_k p_k(\hat{\mu})\, C(\hat{\mu}_k, s_k(t)) / m_P(b)$ on each pool; if it exceeds the theory's $\kappa_q$ bound at $c = 0.4$, the decoded subset is not quantization-faithful and the minimax-rate conclusion does not fo

Watch

Extended reading notes

Core claim

The paper's central claim is that three seemingly rival cold-start strategies — typicality (TypiClust), coverage (ProbCover), and diversity (ActiveFT) — are exact realizations of one transport-selection program, differing only in representatives, cost, and decoder. Its balanced entropic branch defines $\varepsilon$-AS, whose only tunable quantity is the regularization strength $\varepsilon$. The paper proves that $\varepsilon^* = c\,\hat{m}(b)$, a dimensionless constant times the mean closest-anchor cost, is the unique scale-equivariant rule, and that it attains the task-agnostic minimax rate $\Theta(b^{-2/d_{\mathrm{int}}})$: label-optimal as budget $b$ grows, with the exponent set by the m

Load-bearing premise

The theory certifies the minimax rate only for a calibrated constant $c$ inside an unspecified window $[t_-, t_+]$; the paper fixes $c = 0.4$ by one grid search on MNIST labels and never verifies that $0.4$ lies inside that window on any other dataset or budget.

Editorial extensions

If this is right

  • A user no longer needs to choose among typicality, coverage, or diversity: $\varepsilon$-AS derives the balance from the unlabeled pool and the budget alone.
  • Label efficiency is governed by the feature manifold's intrinsic dimension: minimax risk decays as $b^{-2/d_{\mathrm{int}}}$, so pools with larger intrinsic dimension need larger budgets for the same target error.
  • The budget-dependent error floor is characterized exactly: no task-agnostic selector can beat $\tfrac{L^2}{2} m_P(b)$, and the floor is attainable by any quantization-faithful selector.
  • The scale rule $\varepsilon^* = c\hat{m}(b)$ is the unique statistic consistent with the paper's axioms, so future selectors built inside the same framework inherit the same adaptive temperature.
  • On ImageNet-1k, $\varepsilon$-AS beats the strongest fixed-rule baseline by 1.29% average accuracy while cutting selection time from 55.4 to 24.3 seconds (2.3$\times$).

Reading between the lines

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

  • If $c$ is truly universal, the same frozen constant should transfer to other frozen feature extractors without recalibration, as long as the pool geometry has a well-defined intrinsic dimension — a direct test the paper does not run.
  • The axioms that single out the mean closest-anchor cost explicitly rule out robust statistics such as the median or trimmed mean; on pools with heavy-tailed closest-anchor costs, comparing $\varepsilon^* = c\hat{m}(b)$ against a median-based rule would probe whether the mean is load-bearing.
  • The unification relocates method design to the decoder: any new decoding rule inside the same balanced entropic branch inherits the proven minimax rate, which could yield cheaper or parallelizable selection schemes.
  • Extending the canonical two-route calibration game with a third route, such as a coverage-fidelity term, would produce a calibration surface $c(A, \lambda, \kappa)$ that could replace the single MNIST-fixed constant and make the one-knob claim fully data-driven.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This manuscript proposes a unified optimal-transport (OT) view of cold-start active learning (CSAL). It defines a transport-selection realization that exactly subsumes TypiClust, ProbCover, and ActiveFT (Theorem 1), establishes a task-agnostic minimax bound governed by the closest-anchor quantization cost mP(b) (Theorem 2), and derives a Sinkhorn-based selector, ε-AS, with data-adaptive temperature ε* = c·bm(b) (Theorem 3). Experiments on MNIST, CIFAR-10/100, STL-10, Caltech-101, and ImageNet-1k report the highest point estimate in 25 of 26 settings, a 1.29% average improvement over ActiveFT on ImageNet-1k, and a 2.3× selection-time speedup.

Significance. The unification is conceptually valuable: it reframes the choice among typicality/coverage/diversity methods as the choice of a single allocation scale, and the minimax connection to mP(b) gives a principled reason to adapt that scale to the unlabeled pool. The finite-pool subsumption identities are exact and carefully documented, and the appendix is unusually complete. However, the headline claims are not yet fully supported: a supporting inequality in Lemma D.5 is false as stated, the rate guarantee for ε-AS requires c to lie in an unspecified window [t−, t+] that is never verified for any target dataset, and the empirical superiority claim rests on three-seed point estimates with overlapping error bars. These issues are correctable, but they affect the paper's central claims.

major comments (3)
  1. [Appendix D.3, Lemma D.5] The proof of the anchor-to-query transfer bound uses d(z,s_k)^2 ≤ d(z,µ_k)^2 + d(µ_k,s_k)^2. This is false in general Euclidean geometry (e.g., collinear points with both distances equal to 1 give left side 4 and right side 2). The correct inequality is d(z,s_k)^2 ≤ 2d(z,µ_k)^2 + 2d(µ_k,s_k)^2, which changes Lemma D.5 to QP(S) ≤ 4RP(µ) + 4Σ pk C(µk,sk), and the decoder-fidelity constants in Theorem 2 (κq > 2(1+κd)) become κq > 4(1+κd). The minimax rate survives, but the proof as written is incorrect and should be corrected.
  2. [Theorem 3 and Assumption B.6 (Appendix D.4)] The main-text Theorem 3 states that ε-AS attains the minimax rate 'with decoder fidelity,' omitting the appendix's condition that c must lie in the normalized-temperature window [t−, t+] of Assumption B.6. The experimental protocol selects c = 0.4 by grid search on MNIST labels (Section VI, Calibration) and never checks whether 0.4 lies in [t−, t+] on CIFAR-10, CIFAR-100, STL-10, Caltech-101, or ImageNet-1k. Since the canonical calibration game's ccan depends on dimensionless weights (A, λ, κ) that are not tied to data, the universality of c = 0.4 is not established. Please add a label-free diagnostic of the Assumption B.6 window (e.g., estimate supt QP(S_t bm)/mP(b) per dataset and budget) or weaken the claim to 'conditional on c in the theoretical window.'
  3. [Section VI, Tables E.3–E.8] The headline '25 of 26 settings' is based on three-seed point estimates with widely overlapping standard deviations (e.g., STL-10 b=200: ε-AS 97.48±0.88 vs. ActiveFT 97.03±1.06; CIFAR-100 b=100: 69.77±1.66 vs. ActiveFT 67.76±1.46). No significance test or paired bootstrap is reported, so 'consistently achieves state-of-the-art' is not statistically supported as written. In addition, the c-sensitivity analysis (Figure 3) is limited to STL-10 and Caltech-101; the stability of the transferred constant on CIFAR-10/100 and ImageNet-1k is not demonstrated.
minor comments (4)
  1. [Algorithm 1, line 4] εmin is an absolute floor in the algorithm, but exact cost-scale equivariance requires the floor to be rescaled together with the cost. Please specify in Algorithm 1 that εmin should scale with bm(b), or state the asymptotic condition εmin = o(mP(b)) in the algorithm's input description.
  2. [Section V, Eq. (7)] The 'mixture consistency' axiom of Definition D.6 applies only to cost lists evaluated under a common, fixed anchor and cost system. This qualification appears only in Appendix D.4; it should be stated where the rule is introduced in the main text.
  3. [Figure 2 and Tables E.3–E.8] The number of seeds (three) is stated in the experimental setup but not in the figure caption or the main tables. Please add it to the caption/table headings for clarity.
  4. [Theorem 1 statement] The 'exact subsumption' claim is precise only for the evaluated specializations (e.g., deterministic no-repeat ProbCover and K=b TypiClust). Please state this qualification directly in the theorem statement, as it is currently only implicit in the appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the main derivation chain is self-contained; the calibrated constant is a hyperparameter transfer and the theoretical guarantees are explicitly conditional on stated assumptions.

full rationale

The paper's core theoretical chain is not circular. Theorem 1 is an exact finite-pool algebraic identification: each realization is parameterized so that minimizing out the transport plan algebraically recovers the method's objective (Appendix D.1), and the proof constructs the reductions explicitly. Theorem 2 is a genuine minimax result: the lower bound uses two indistinguishable Lipschitz tasks f0=0 and f1=L*d(.,S) that agree on every queried point, and the upper bound uses nearest-neighbor readout under the stated quantization condition; both bounds are proved in D.3. Theorem 3's consistency statement is a plug-in result: epsilon*=c*\m(b) inherits \m(b)/m_P(b)->1 from Assumptions B.4-B.5. The 'unique' temperature statistic is an axiomatic characterization (Definition D.6 + Lemma D.7); the mixture-consistency axiom indeed largely encodes the weighted arithmetic mean, but the paper states this axiom explicitly and the proof is a short induction, so it is a transparent design assumption rather than a hidden circular reduction. The empirical SOTA claim is an external transfer evaluation: c=0.4 is selected once on MNIST labels and then frozen; CIFAR-10/100, STL-10, Caltech-101, and ImageNet-1k contribute no labels to choosing c. The minimax-rate guarantee for epsilon-AS is explicitly conditional on the unverified window c in [t_-,t_+] and Assumption B.6 (Appendix D.4, Table A.1); the paper itself lists 'eliminating the remaining global calibration' as future work (Section VI Discussion). That is a correctness/robustness gap, not circularity. No load-bearing result is imported from the authors' prior work: the minimax, consistency, and uniqueness theorems are proved in this paper. Therefore no circular step meeting the hard-evidence standard is present.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The central theory rests on standard compact-sphere assumptions plus a density assumption (B.6) on the decoded queries; the only fitted scalar is c=0.4. The 'uniqueness' of the temperature rule follows from axioms chosen to imply the arithmetic mean, so that part of the derivation is somewhat self-referential but not circular in the sense of assuming the target result.

free parameters (2)
  • c (dimensionless regularization multiplier) = 0.4
    Multiplier in epsilon_star = c * m_hat(b); chosen by grid search on MNIST labels and transferred to all other datasets and budgets; not derived from the theory.
  • epsilon_min numerical floor = 1e-4
    Entropic floor for numerical stability; asymptotically required to be o(m_P(b)); not a scientific parameter.
assumptions (7)
  • domain assumption Assumption B.1: normalized features supported on the unit sphere, distribution not supported on at most b points
    Defines the geometry; invoked in Theorem 2 and Theorem 3.
  • domain assumption Assumption B.2: intrinsic-dimensional scaling m_P(b) = Theta(b^(-2/d_int))
    Needed to translate the minimax bound into a rate in b; the paper fits this in Appendix E.6.
  • standard math Assumption B.3: i.i.d. pool sampling and independent clustering initialization
    Standard statistical model for the unlabeled pool.
  • domain assumption Assumption B.4: uniform empirical approximation error r_n,b = o(m_P(b))
    Ensures the empirical objective tracks the population objective; plausible on compact spheres.
  • domain assumption Assumption B.5: approximated clustering gap xi_n,b = o(m_P(b))
    Requires the k-means solver to be near-optimal relative to the resolution scale.
  • domain assumption Assumption B.6: round-robin decoder places each anchor's query within distance small relative to m_P(b), uniformly over t in [t_minus, t_plus]
    The load-bearing assumption for the decoded minimax rate; only justified heuristically by 'dense pool', not verified on real data.
  • ad hoc to paper Scale-rule axioms in Definition D.6: permutation, positive-scale, mixture consistency, normalization
    These four axioms force the unique form c * m_hat(b); the mixture-consistency axiom in particular is a modeling choice that directly builds in the arithmetic mean.

how reviews work

0 comments
Cite this review

Pith. "Pith review of One Knob to Rule Them All: A Unified Optimal Transport View of Cold-Start Active Learning." pith.science (2026). https://pith.science/paper/EQZTJM4Y

@misc{pith2026260803249,
  author       = {Pith},
  title        = {Pith review of: One Knob to Rule Them All: A Unified Optimal Transport View of Cold-Start Active Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EQZTJM4Y}},
  note         = {Machine review of arXiv:2608.03249}
}
abstract

Cold-Start Active Learning (CSAL) aims to select a valuable subset from an unlabeled pool without any prior knowledge or human assistance. Existing methods take diverse routes based on typicality, coverage, or diversity. Each rests on its own inductive bias and therefore performs well on some tasks yet poorly on others. We argue that the real challenge is not to design yet another selection heuristic, but to make CSAL adapt automatically to the data and task at hand. To this end, we revisit CSAL through the lens of optimal transport. First, we propose a generalized transport selection framework that reveals the shared allocation structure of existing methods and exactly subsumes representative formulations. Second, we introduce a theoretical analysis that characterizes the trade-off controlled by entropic regularization and establishes a task-agnostic minimax bound for cold-start selection. These results provide a principled foundation for adapting the regularization strength to the unlabeled data. Third, we derive a data-adaptive regularization rule and present a novel Sinkhorn-based CSAL algorithm, termed $\epsilon$-Adaptive Selection ($\epsilon$-AS). Extensive experiments on six public datasets and multiple annotation budgets show that $\epsilon$-AS consistently achieves state-of-the-art performance. On ImageNet-1k, it improves the average accuracy over ActiveFT by 1.29% while reducing selection time by 56.2%. Code will be released at https://github.com/Z-yiwei/OT-CSAL

Figures

Figures reproduced from arXiv: 2608.03249 by the authors.

Figure 1
Figure 1. Fixed geometric selection rules do not transfer consistently across datasets. Each cell gives one method’s accuracy rank on a dataset. Baseline ranks vary greatly while ε-AS ranks first on all six datasets. repeated human interaction from the acquisition process, but it also removes the task feedback on which warm-start meth￾ods rely. Therefore, the learner must instead make its de￾cision from the geometry of pretra… view at source ↗
Figure 2
Figure 2. Top-1 accuracy on six benchmarks. ε-AS leads in 25 of 26 dataset and budget settings. Main Results [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Sensitivity to the calibration constant and Sinkhorn iteration cap. The dark curve reports their mean, and the shaded band spans the budget-level range. The figure shows changes in top-1 accuracy relative to c = 0.4 and Tsk = 200. ActiveFT by 1.29%. It also reduces selection time from 55.41 to 24.28 seconds, which gives a 2.3× speedup. DCoM and USL require more time and achieve lower accuracy. Typi￾Clust is faster, … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Accuracy–efficiency trade-off on ImageNet-1k averaged over annotation budgets and seeds. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

77 extracted references · 73 canonical work pages

  1. [1]

    Sinkhorn Distances: Lightspeed Computation of Optimal Transport

    Cuturi, Marco. Sinkhorn Distances: Lightspeed Computation of Optimal Transport. Advances in Neural Information Processing Systems

    work page

  2. [2]

    Convergence of Entropic Schemes for Optimal Transport and Gradient Flows

    Carlier, Guillaume and Duval, Vincent and Peyr \'e , Gabriel and Schmitzer, Bernhard. Convergence of Entropic Schemes for Optimal Transport and Gradient Flows. SIAM Journal on Mathematical Analysis

    work page

  3. [3]

    A Survey of the Schr \"o dinger Problem and Some of Its Connections with Optimal Transport

    L \'e onard, Christian. A Survey of the Schr \"o dinger Problem and Some of Its Connections with Optimal Transport. Discrete and Continuous Dynamical Systems

    work page

  4. [4]

    Sample Complexity of Sinkhorn Divergences

    Genevay, Aude and Chizat, L \'e na \"i c and Bach, Francis and Cuturi, Marco and Peyr \'e , Gabriel. Sample Complexity of Sinkhorn Divergences. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics

    work page

  5. [5]

    Statistical Bounds for Entropic Optimal Transport: Sample Complexity and the Central Limit Theorem

    Mena, Gonzalo and Niles-Weed, Jonathan. Statistical Bounds for Entropic Optimal Transport: Sample Complexity and the Central Limit Theorem. Advances in Neural Information Processing Systems

    work page

  6. [6]

    Interpolating between Optimal Transport and MMD using Sinkhorn Divergences

    Feydy, Jean and S \'e journ \'e , Thibault and Vialard, Fran c ois-Xavier and Amari, Shun-ichi and Trouv \'e , Alain and Peyr \'e , Gabriel. Interpolating between Optimal Transport and MMD using Sinkhorn Divergences. Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics

    work page

  7. [7]

    Computational Optimal Transport: with Applications to Data Sciences

    Peyr \'e , Gabriel and Cuturi, Marco. Computational Optimal Transport: with Applications to Data Sciences. Foundations and Trends in Machine Learning

    work page

  8. [8]

    Optimal Transport: Old and New

    Villani, C \'e dric. Optimal Transport: Old and New

    work page

Show all 77 references
  1. [9]

    Near-Linear Time Approximation Algorithms for Optimal Transport via Sinkhorn Iteration

    Altschuler, Jason and Niles-Weed, Jonathan and Rigollet, Philippe. Near-Linear Time Approximation Algorithms for Optimal Transport via Sinkhorn Iteration. Advances in Neural Information Processing Systems

  2. [10]

    Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets

    Hacohen, Guy and Dekel, Avihu and Weinshall, Daphna. Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets. Proceedings of the 39th International Conference on Machine Learning (ICML)

  3. [11]

    Active Learning Through a Covering Lens

    Yehuda, Ofer and Dekel, Avihu and Hacohen, Guy and Weinshall, Daphna. Active Learning Through a Covering Lens. Advances in Neural Information Processing Systems

  4. [12]

    Active Finetuning: Exploiting Annotation Budget in the Pretraining-Finetuning Paradigm

    Xie, Yichen and Lu, Han and Yan, Junchi and Yang, Xiaokang and Tomizuka, Masayoshi and Zhan, Wei. Active Finetuning: Exploiting Annotation Budget in the Pretraining-Finetuning Paradigm. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2023

  5. [13]

    DCoM: Active Learning for All Learners

    Mishal, Inbal and Weinshall, Daphna. DCoM: Active Learning for All Learners. arXiv preprint arXiv:2407.01804

  6. [14]

    Unsupervised Selective Labeling for More Effective Semi-Supervised Learning

    Wang, Xudong and Lian, Long and Yu, Stella X. Unsupervised Selective Labeling for More Effective Semi-Supervised Learning. European Conference on Computer Vision (ECCV)

  7. [15]

    Active Learning for Convolutional Neural Networks: A Core-Set Approach

    Sener, Ozan and Savarese, Silvio. Active Learning for Convolutional Neural Networks: A Core-Set Approach. International Conference on Learning Representations (ICLR)

  8. [16]

    and Zhang, Chicheng and Krishnamurthy, Akshay and Langford, John and Agarwal, Alekh

    Ash, Jordan T. and Zhang, Chicheng and Krishnamurthy, Akshay and Langford, John and Agarwal, Alekh. Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds. International Conference on Learning Representations (ICLR)

  9. [17]

    Low-Budget Active Learning via Wasserstein Distance: An Integer Programming Approach

    Mahmood, Rafid and Fidler, Sanja and Law, Marc T. Low-Budget Active Learning via Wasserstein Distance: An Integer Programming Approach. International Conference on Learning Representations (ICLR)

  10. [18]

    and Zhou, Zongwei

    Chen, Liangyu and Bai, Yutong and Huang, Siyu and Lu, Yongyi and Wen, Bihan and Yuille, Alan L. and Zhou, Zongwei. Making Your First Choice: To Address Cold Start Problem in Medical Active Learning. Medical Imaging with Deep Learning

  11. [19]

    Parting with Illusions about Deep Active Learning

    Mittal, Sudhanshu and Tatarchenko, Maxim and C i c ek, \"O zg \"u n and Brox, Thomas. Parting with Illusions about Deep Active Learning. arXiv preprint arXiv:1912.05361

    1912 arXiv

  12. [20]

    Towards Robust and Reproducible Active Learning Using Neural Networks

    Munjal, Prateek and Hayat, Nasir and Hayat, Munawar and Sourati, Jamshid and Khan, Shadab. Towards Robust and Reproducible Active Learning Using Neural Networks. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2022

  13. [21]

    Active Learning Literature Survey

    Settles, Burr. Active Learning Literature Survey

  14. [22]

    Deep Bayesian Active Learning with Image Data

    Gal, Yarin and Islam, Riashat and Ghahramani, Zoubin. Deep Bayesian Active Learning with Image Data. Proceedings of the 34th International Conference on Machine Learning (ICML)

  15. [23]

    Learning Algorithms for Active Learning

    Bachman, Philip and Sordoni, Alessandro and Trischler, Adam. Learning Algorithms for Active Learning. Proceedings of the 34th International Conference on Machine Learning

  16. [24]

    2025 , author =

    Deep adaptive wavelet autoencoder with mutually independent empirical cumulative distribution for unsupervised motor anomaly detection , journal =. 2025 , author =

    2025

  17. [25]

    High-Order Flow Matching: Unified Framework and Sharp Statistical Rates , volume =

    Su, Maojiang and Hu, Jerry Yao-Chieh and Lee, Yi-Chen and Zhu, Ning and Chung, Jui-Hui and Wu, Shang and Song, Zhao and Chen, Minshuo and Liu, Han , booktitle =. High-Order Flow Matching: Unified Framework and Sharp Statistical Rates , volume =

  18. [26]

    BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning

    Kirsch, Andreas and van Amersfoort, Joost and Gal, Yarin. BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning. Advances in Neural Information Processing Systems

  19. [27]

    and Goel, Surbhi and Krishnamurthy, Akshay and Kakade, Sham

    Ash, Jordan T. and Goel, Surbhi and Krishnamurthy, Akshay and Kakade, Sham. Gone Fishing: Neural Active Learning with Fisher Embeddings. Advances in Neural Information Processing Systems

  20. [28]

    Batch Active Learning at Scale

    Citovsky, Gui and DeSalvo, Giulia and Gentile, Claudio and Karydas, Lazaros and Rajagopalan, Anand and Rostamizadeh, Afshin and Kumar, Sanjiv. Batch Active Learning at Scale. Advances in Neural Information Processing Systems

  21. [29]

    Selection via Proxy: Efficient Data Selection for Deep Learning

    Coleman, Cody and Yeh, Christopher and Mussmann, Stephen and Mirzasoleiman, Baharan and Bailis, Peter and Liang, Percy and Leskovec, Jure and Zaharia, Matei. Selection via Proxy: Efficient Data Selection for Deep Learning. International Conference on Learning Representations

  22. [30]

    and Gale, William A

    Lewis, David D. and Gale, William A. A Sequential Algorithm for Training Text Classifiers. Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval

  23. [31]

    Active Hidden Markov Models for Information Extraction

    Scheffer, Tobias and Decomain, Christian and Wrobel, Stefan. Active Hidden Markov Models for Information Extraction. International Symposium on Intelligent Data Analysis (IDA)

  24. [32]

    A Mathematical Theory of Communication

    Shannon, Claude E. A Mathematical Theory of Communication. Bell System Technical Journal

  25. [33]

    u rnberger, Andreas and K \

    Beluch, William H. and Genewein, Tim and N \"u rnberger, Andreas and K \"o hler, Jan M. The Power of Ensembles for Active Learning in Image Classification. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition

    2018

  26. [34]

    Unsupervised Anomaly Detection for Aircraft PRSOV With Random Projection-Based Inner Product Prediction , year=

    Peng, Dandan and Zhu, Ning and Han, Te and Chen, Zhuyun and Liu, Chenyu , journal=. Unsupervised Anomaly Detection for Aircraft PRSOV With Random Projection-Based Inner Product Prediction , year=

  27. [35]

    Learning Loss for Active Learning

    Yoo, Donggeun and Kweon, In So. Learning Loss for Active Learning. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2019

  28. [36]

    Variational Adversarial Active Learning

    Sinha, Samarth and Ebrahimi, Sayna and Darrell, Trevor. Variational Adversarial Active Learning. 2019 IEEE/CVF International Conference on Computer Vision (ICCV)

    2019

  29. [37]

    and Chen, Xiaojiang and Wang, Xin

    Ren, Pengzhen and Xiao, Yun and Chang, Xiaojun and Huang, Po-Yao and Li, Zhihui and Gupta, Brij B. and Chen, Xiaojiang and Wang, Xin. A Survey of Deep Active Learning. ACM Computing Surveys

  30. [38]

    An Adversarial Training Framework Based on Unsupervised Feature Reconstruction Constraints for Crystalline Silicon Solar Cells Anomaly Detection , year=

    Zhu, Ning and Wang, Jing and Zhang, Ying and Wang, Huan and Han, Te , journal=. An Adversarial Training Framework Based on Unsupervised Feature Reconstruction Constraints for Crystalline Silicon Solar Cells Anomaly Detection , year=

  31. [39]

    Adversarial Frequency Component Reconstruction Constraint for Helicopter Vibration Signal Anomaly Detection: An Unsupervised Dual-Domain Approach , year=

    Zhu, Ning and Dong, Tianzhi Xu and Peng, Dandan , journal=. Adversarial Frequency Component Reconstruction Constraint for Helicopter Vibration Signal Anomaly Detection: An Unsupervised Dual-Domain Approach , year=

  32. [40]

    Contextual Diversity for Active Learning

    Agarwal, Sharat and Arora, Himanshu and Anand, Saket and Arora, Chetan. Contextual Diversity for Active Learning. European Conference on Computer Vision (ECCV)

  33. [41]

    Submodularity in Data Subset Selection and Active Learning

    Wei, Kai and Iyer, Rishabh and Bilmes, Jeff. Submodularity in Data Subset Selection and Active Learning. Proceedings of the 32nd International Conference on Machine Learning

  34. [42]

    Efficient Coreset Constructions via Sensitivity Sampling

    Braverman, Vladimir and Feldman, Dan and Lang, Harry and Statman, Adiel and Zhou, Samson. Efficient Coreset Constructions via Sensitivity Sampling. Proceedings of the 13th Asian Conference on Machine Learning

  35. [43]

    Coresets for Estimating Means and Mean Square Error with Limited Greedy Samples

    Vahidian, Saeed and Mirzasoleiman, Baharan and Cloninger, Alexander. Coresets for Estimating Means and Mean Square Error with Limited Greedy Samples. Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence

  36. [44]

    Mind the Boundary: Coreset Selection via Reconstructing the Decision Boundary

    Yang, Shuo and Cao, Zhe and Guo, Sheng and Zhang, Ruiheng and Luo, Ping and Zhang, Shengping and Nie, Liqiang. Mind the Boundary: Coreset Selection via Reconstructing the Decision Boundary. Proceedings of the 41st International Conference on Machine Learning

  37. [45]

    Foundation Model Makes Clustering A Better Initialization for Cold-Start Active Learning

    Yuan, Han and Hong, Chuan. Foundation Model Makes Clustering A Better Initialization for Cold-Start Active Learning. arXiv preprint arXiv:2402.02561

    arXiv

  38. [46]

    Cold-start Active Learning through Self-supervised Language Modeling

    Yuan, Michelle and Lin, Hsuan-Tien and Boyd-Graber, Jordan. Cold-start Active Learning through Self-supervised Language Modeling. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

    2020

  39. [47]

    Reducing Label Effort: Self-Supervised Meets Active Learning

    Bengar, Javad Zolfaghari and van de Weijer, Joost and Twardowski, Bartlomiej and Raducanu, Bogdan. Reducing Label Effort: Self-Supervised Meets Active Learning. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)

  40. [48]

    Generalized Coverage for More Robust Low-Budget Active Learning

    Bae, Wonho and Noh, Junhyug and Sutherland, Danica J. Generalized Coverage for More Robust Low-Budget Active Learning. European Conference on Computer Vision (ECCV)

  41. [49]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    SUGFW: A SAM-Based Uncertainty-Guided Feature Weighting Framework for Cold Start Active Learning , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

  42. [50]

    and Oliveira, Gabriel L

    Bae, Wonho and Sutherland, Danica J. and Oliveira, Gabriel L. Uncertainty Herding: One Active Learning Method for All Label Budgets. International Conference on Learning Representations (ICLR)

  43. [51]

    Guo, Jiaxin and Chen, C. L. Philip and Li, Shuzhen and Zhang, Tong. DEUCE: Dual-diversity Enhancement and Uncertainty-awareness for Cold-start Active Learning. Transactions of the Association for Computational Linguistics (TACL)

  44. [52]

    CSAL-3D: Cold-Start Active Learning for 3D Medical Image Segmentation via SSL-Driven Uncertainty-Reinforced Diversity Sampling

    Zhu, Ning and Ye, Ping and Zhong, Lanfeng and Yue, Qiang and Zhang, Shaoting and Wang, Guotai. CSAL-3D: Cold-Start Active Learning for 3D Medical Image Segmentation via SSL-Driven Uncertainty-Reinforced Diversity Sampling. Medical Image Computing and Computer Assisted Interven...

  45. [53]

    SUGFW: A SAM-Based Uncertainty-Guided Feature Weighting Framework for Cold-Start Active Learning

    Ma, Xiaochuan and Fu, Jia and Zhong, Lanfeng and Zhu, Ning and Wang, Guotai. SUGFW: A SAM-Based Uncertainty-Guided Feature Weighting Framework for Cold-Start Active Learning. Medical Image Computing and Computer Assisted Intervention (MICCAI)

  46. [54]

    Active Learning for Vision-Language Models

    Safaei, Bardia and Patel, Vishal M. Active Learning for Vision-Language Models. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

  47. [55]

    Transactions on Machine Learning Research

    Maxime Oquab and Timothée Darcet and Théo Moutakanni and Huy Vo and Marc Szafraniec and Vasil Khalidov and Pierre Fernandez and Daniel Haziza and Francisco Massa and Alaaeldin El-Nouby and Mahmoud Assran and Nicolas Ballas and Wojciech Galuba and Russell Howes and Po-Yao Huang...

  48. [56]

    Sigmoid Loss for Language Image Pre-Training

    Zhai, Xiaohua and Mustafa, Basil and Kolesnikov, Alexander and Beyer, Lucas. Sigmoid Loss for Language Image Pre-Training. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

  49. [57]

    Gradient-Based Learning Applied to Document Recognition

    LeCun, Yann and Bottou, L \'e on and Bengio, Yoshua and Haffner, Patrick. Gradient-Based Learning Applied to Document Recognition. Proceedings of the IEEE

  50. [58]

    Learning Multiple Layers of Features from Tiny Images

    Krizhevsky, Alex. Learning Multiple Layers of Features from Tiny Images

  51. [59]

    An Analysis of Single-Layer Networks in Unsupervised Feature Learning

    Coates, Adam and Ng, Andrew and Lee, Honglak. An Analysis of Single-Layer Networks in Unsupervised Feature Learning. Proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS)

  52. [60]

    Learning Generative Visual Models from Few Training Examples: An Incremental Bayesian Approach Tested on 101 Object Categories

    Fei-Fei, Li and Fergus, Rob and Perona, Pietro. Learning Generative Visual Models from Few Training Examples: An Incremental Bayesian Approach Tested on 101 Object Categories. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshop

  53. [61]

    ImageNet: A Large-Scale Hierarchical Image Database

    Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li. ImageNet: A Large-Scale Hierarchical Image Database. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

  54. [62]

    Asymptotic Methods in Statistical Decision Theory

    Le Cam, Lucien. Asymptotic Methods in Statistical Decision Theory

  55. [63]

    Introduction to Nonparametric Estimation

    Tsybakov, Alexandre B. Introduction to Nonparametric Estimation

  56. [64]

    Concentration Inequalities: A Nonasymptotic Theory of Independence

    Boucheron, St \'e phane and Lugosi, G \'a bor and Massart, Pascal. Concentration Inequalities: A Nonasymptotic Theory of Independence

  57. [65]

    Maximum Likelihood Estimation of Intrinsic Dimension

    Levina, Elizaveta and Bickel, Peter J. Maximum Likelihood Estimation of Intrinsic Dimension. Advances in Neural Information Processing Systems , year = 2004, volume =

    2004

  58. [66]

    Gamma-Convergence for Beginners

    Braides, Andrea. Gamma-Convergence for Beginners

  59. [67]

    An Introduction to Gamma-Convergence

    Dal Maso, Gianni. An Introduction to Gamma-Convergence

  60. [68]

    One-shot Active Learning for Image Segmentation via Contrastive Learning and Diversity-based Sampling

    Jin, Qiuye and Yuan, Mingzhi and Qiao, Qin and Song, Zhijian. One-shot Active Learning for Image Segmentation via Contrastive Learning and Diversity-based Sampling. Knowledge-Based Systems

  61. [69]

    Least Squares Quantization in PCM

    Lloyd, Stuart P. Least Squares Quantization in PCM. IEEE Transactions on Information Theory

  62. [70]

    Billion-scale Similarity Search with GPUs

    Johnson, Jeff and Douze, Matthijs and J \'e gou, Herv \'e. Billion-scale Similarity Search with GPUs. IEEE Transactions on Big Data

  63. [71]

    Cold-Start Active Learning for Image Classification

    Jin, Qiuye and Yuan, Mingzhi and Li, Shiman and Wang, Haoran and Wang, Manning and Song, Zhijian. Cold-Start Active Learning for Image Classification. Information Sciences

  64. [72]

    and Xu, Jianliang

    Cao, Xiaofeng and Tsang, Ivor W. and Xu, Jianliang. Cold-Start Active Sampling Via -Tube. IEEE Transactions on Cybernetics

  65. [73]

    An Efficient Query Strategy for Active Learning via Optimal Transport

    Zhu, Zhenyu and Zhang, Teng and Shi, Xuanhua and Jin, Hai. An Efficient Query Strategy for Active Learning via Optimal Transport

  66. [74]

    Partial Optimal Transport for Support Subset Selection

    Riaz, Bilal and Karahan, Y \"u ksel and Brockmeier, Austin J. Partial Optimal Transport for Support Subset Selection. Transactions on Machine Learning Research

  67. [75]

    and Ramakrishnan, Ganesh and Mishra, Bamdev and Jawanpuria, Pratik

    Chanda, Prateek and Agrawal, Prayas and Gurumoorthy, Karthik S. and Ramakrishnan, Ganesh and Mishra, Bamdev and Jawanpuria, Pratik. UniPROT: Uniform Prototype Selection via Partial Optimal Transport with Submodular Guarantees. Proceedings of the 29th International Conference o...

  68. [76]

    Dataset-Aware Cold-Start Active Learning for Annotation-Efficient 3D Medical Image Segmentation

    Hattat, R \'e mi and Beaumont, Marine and Bertholdt, Charline and Hossu, Gabriela and Morel, Olivier and Chen, Bailiang. Dataset-Aware Cold-Start Active Learning for Annotation-Efficient 3D Medical Image Segmentation. arXiv preprint arXiv:2606.20765

    arXiv

  69. [77]

    MedCAL-Bench: A Comprehensive Benchmark on Cold-Start Active Learning with Foundation Models for Medical Image Analysis

    Zhu, Ning and Ma, Xiaochuan and Zhang, Shaoting and Wang, Guotai. MedCAL-Bench: A Comprehensive Benchmark on Cold-Start Active Learning with Foundation Models for Medical Image Analysis. arXiv preprint arXiv:2508.03441

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.