Pith. sign in

REVIEW 4 major objections 5 minor 27 references

A new online method, ARROW, reweights each minibatch to a moving reference to stabilize MMD and CORAL domain adaptation on streaming data.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 09:59 UTC pith:RVW7SZ7Z

load-bearing objection A plausible online variance-reduction method for MMD/CORAL with solid experiments, but the core mechanism is asserted rather than proven and the streaming setting is not actually tested. the 4 major comments →

arxiv 2607.20374 v1 pith:RVW7SZ7Z submitted 2026-07-22 cs.LG

Online Variance Reduction for Domain Adaptation on Streaming Data

Andrea Napoli This is my paper
classification cs.LG
keywords online variance reductiondomain adaptationmaximum mean discrepancyCORALinstance reweightingexponential moving averagestreaming datastochastic optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper claims that the noisy gradients of the MMD and CORAL domain-adaptation losses can be variance-reduced in a purely online, streaming setting, something previous variance-reduction methods could not do because they required access to the whole dataset. The proposed ARROW method keeps an exponentially weighted moving average of the alignment statistics and, at each step, chooses per-sample weights that pull the current minibatch's statistic toward that reference. Because the weights are unconstrained, the weight-solving step becomes a convex quadratic program for MMD and a linear least-squares problem for CORAL, both solvable with standard techniques. The author argues this makes variance-reduced domain adaptation feasible for online, distributed, or incremental learning, and reports that ARROW matches offline methods in variance reduction, speed, and target accuracy.

Core claim

On its own terms, the central claim is that stochastic variance reduction for the MMD and CORAL losses does not require finite-sum structure or offline access to the data. It can be achieved by replacing the unknown true alignment difference D with an exponentially weighted moving-average estimate \tilde{D} and, at each timestep, solving a reweighting problem that aligns the weighted minibatch statistic to that estimate. For MMD this is a convex quadratic in the sample weights; for CORAL, after a relaxation that fixes the covariance centres, it becomes a linear least-squares problem. The paper reports that this online scheme delivers variance reduction and target-domain accuracy competitive

What carries the argument

The central mechanism is instance reweighting of the minibatch alignment statistic: each incoming sample receives an arbitrary (not necessarily convex) weight so that the weighted kernel-mean difference (for MMD) or weighted covariance difference (for CORAL) matches a running EWMA reference statistic. The reference tracks the true population statistic over time, and the per-step weight optimisation is a convex quadratic program (MMD) or a linear least-squares problem (CORAL) that can be solved with standard linear-algebra routines.

Load-bearing premise

The load-bearing premise is that aligning the weighted minibatch statistic to the EWMA reference faithfully proxies for reducing the true discrepancy-gradient error; if the EWMA lags behind a distribution shift, or if unconstrained weights distort the gradient, the claimed variance reduction is not guaranteed.

What would settle it

Run ARROW on a stream with an abrupt mid-stream distribution shift, and measure the actual gradient error \|\hat{\nabla}L - \nabla L\| at each step. If the error spikes sharply after the shift and does not recover, the EWMA reference lags too much. A second check is to replace the EWMA with an oracle that has access to the true population statistic: if ARROW's variance reduction is much worse than the oracle's, the reference, not the reweighting, is the bottleneck.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If ARROW is correct, domain-adaptation training can be stabilised on streaming or unbounded data without storing or revisiting past samples.
  • The variance-reduction step's runtime depends only on the minibatch size and a small buffer of historical minibatches, not on dataset size, making it viable for extremely large or infinite streams.
  • The same reweighting-to-a-moving-reference mechanism may extend to other non-additive losses beyond MMD and CORAL.
  • Because the reference is updated continuously, the method can track time-varying distributions, a capability offline SVR methods lack.
  • In the reported experiments, integrating ARROW into MMD/CORAL training yields target accuracy competitive with substantially more complex domain-adaptation methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural extension would be to swap the EWMA for other online estimators with decreasing step sizes, which might give provable convergence guarantees for the reweighting scheme.
  • Since the weights are unconstrained, they could overfit a single minibatch to the reference and inflate gradient norms; adding weight regularisation or clipping is a concrete, testable variant.
  • ARROW could be combined with within-minibatch variance-reduction tricks, such as antithetic pairing, to further lower gradient noise without needing full-dataset access.
  • A direct measurement of true gradient error, rather than the proxy such as the distance between the weighted statistic and the EWMA, would reveal exactly when the EWMA proxy fails, such as under abrupt domain shifts.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces ARROW, an online stochastic variance reduction method for the MMD and CORAL domain-adaptation losses. ARROW maintains an exponentially weighted moving average (EWMA) of the discrepancy statistic and, at each minibatch, solves a quadratic program that reweights the current minibatch so that its weighted discrepancy statistic matches the EWMA reference. For CORAL, a relaxation using linear combinations of outer products is proposed. The paper reports Monte Carlo variance comparisons, DomainBed accuracy experiments on Spawrious, Office-Home, and Humpbacks, and wall-clock runtimes. The authors claim that ARROW is the first online SVR method for MMD/CORAL and that it is competitive with offline SVR methods in variance reduction, accuracy, and speed.

Significance. If the central mechanism were proven, ARROW would fill a genuine gap: existing SVR methods for MMD/CORAL require full-dataset access and are unsuitable for streaming, distributed, or incremental UDA. The algorithmic construction is concrete and the quadratic formulation in Eq. (15) is mathematically well-defined. The experiments cover three benchmarks and compare a broad set of sampling-based SVR baselines. However, the paper's load-bearing premise — that aligning a weighted minibatch statistic to an EWMA reference reduces the true gradient error — is asserted rather than proved, and the empirical evaluation does not exercise the streaming/drift regime that motivates the method. The unconstrained weight optimization also raises stability concerns that are not addressed. The idea is promising and the experimental setup is useful, but the current evidence is insufficient to support the full claim.

major comments (4)
  1. [§2.1, Eq. (4)] The central variance-reduction claim is not established. The paper argues that ||∇̂L_disc − ∇L_disc|| ∝ ||D̂_m − D_m|| and then substitutes the EWMA reference D̃_m for the true D_m. This is a three-step leap: (i) the chain-rule proportionality is only schematic, since the gradient also involves the Jacobian of the feature embedding, which changes with model parameters; (ii) the cited smoothness conditions in Zhao & Zhang (2014) and Liu et al. (2020) concern stratified sampling for finite-sum SGD and do not cover a time-varying reference; (iii) D̃_m is a biased, lagged average of past minibatches computed under past model parameters, and with α=0.1 and a 0.01 coefficient cutoff it retains only about 22 minibatches. Under model or data drift, minimizing ||D̂_w − D̃_m|| can increase ||∇̂L − ∇L|| by moving the estimate toward a stale target. There is no SVRG-style correction term and no boun
  2. [§2.2, Eq. (15)] The unconstrained quadratic program is ill-posed as stated. The Gram matrices K_ss and K_tt can be singular, and since no constraints or regularization are imposed on u and v, the objective can be unbounded below when the linear term has a component in the null space, or the optimal weights can be arbitrarily large. The paper explicitly declines KMM-style simplex constraints, but gives no bound on ||u|| or ||v||. Large weights can inflate the weighted discrepancy gradient and destabilize training, which is precisely the failure mode SVR is meant to prevent. The claim that the weighted statistic can 'perfectly align' with the reference if the reference lies in the linear span does not address the case where it does not, nor the effect of large weights on the downstream model update.
  3. [§2.3, Eqs. (20)–(23)] Two issues affect the CORAL extension. First, the relaxation that replaces covariance estimators with fixed-center linear combinations of outer products has no approximation guarantee: the optimized Σ̂_w is not a proper covariance estimate when u and v are arbitrary, and the paper provides no bound relating the relaxed objective to the true covariance-difference error. The statement that the problem is 'solvable to 0 if the linear span contains D̃_CORAL' is about algebraic span, not about estimating the true D_CORAL. Second, the algorithm is under-specified: after solving Eq. (23), the paper never states how the weights enter the actual model gradient — whether the update uses the gradient of ||D̂_w||², the gradient of ||D̂_w − D̃_m||², or something else. This is essential for the claimed variance-reduction behavior.
  4. [§3, Tables 1–3] The experimental evidence is only partially consistent with the paper's accuracy claims, and it does not test the streaming premise. On Spawrious MMD, ARROW (65.2) is far below ORDERED (76.1) and PSDA (73.7); on Office-Home CORAL, ARROW (49.3) is below PSDA (52.7); only Humpbacks shows a clear advantage for ARROW. The claim that ARROW is 'competitive' is therefore selective. More importantly, all benchmarks are static DomainBed datasets and the Monte Carlo simulation in Figure 1 is stationary Gaussian data; there is no experiment with time-varying source/target distributions or feature drift during training. Since the paper's central motivation is online/streaming SVR, the unique selling point of ARROW is not actually exercised. A streaming benchmark with controlled shifts, or at least a synthetic non-stationary experiment, is needed to support the online claim.
minor comments (5)
  1. [Table 2] Office-Home results report only the average standard error across splits, not per-cell standard errors, unlike Tables 1 and 3. This makes row-level comparisons difficult and should be fixed.
  2. [Figure 1] The variance curves are shown without error bars or confidence bands, even though the simulation uses 1,000 samples. Statements like 'comparable to VaRDASS' would be more convincing with uncertainty estimates.
  3. [§3.3] The statement that ARROW's runtime is 'completely independent of dataset size' is overstated: the per-minibatch cost is independent of dataset size, but total runtime scales with the number of streamed minibatches. The paper reports wall-clock times only for fixed-size offline datasets, so the scaling claim is untested.
  4. [§1, References] Several key prior works are self-citations to manuscripts dated 2026 (Napoli & White 2026a,b; Napoli 2026; Baumann et al. 2026) that may not yet be publicly available. Please mark them as 'in press' or provide arXiv identifiers/availability information, since the novelty claim 'first online SVR' depends on these comparisons.
  5. [Table 1] Minor typographical issue in the column header 'A verage'; also, the random search over minibatch size uses k ∼ 2^Uniform(3,7), so k ranges from 8 to 128; the QP cost for the largest k is not discussed.

Circularity Check

0 steps flagged

No circularity found: ARROW's reweighting objective is defined from an EWMA reference, not from the target quantities it is evaluated against, and the experimental comparisons are independent of the optimization objective.

full rationale

The derivation chain in Section 2.1 substitutes an EWMA reference \tilde D_m for the true time-varying statistic D_m in the objective min ||\hat D_m^w - \tilde D_m||^2. This is an approximation/regularity assumption, not a circular definition: \tilde D_m is updated from unweighted minibatch statistics via Eqs. (5)-(6), while the weight optimization in Eq. (15) is a least-squares projection of \tilde D_m onto the span of the current minibatch. The paper's headline claim is about reducing true stochastic gradient error, and it is evaluated externally: Monte Carlo E||\hat D - D||^2 in Figure 1 and target-domain accuracy on DomainBed benchmarks in Tables 1-3, not by the same objective (4) being optimized. The self-citations to Napoli & White / Napoli (VaRDASS, ORDERED, PSDA, etc.) serve as motivation, prior-art context, and baselines; none is used to justify the ARROW construction itself, and no uniqueness theorem or ansatz is imported from those papers. The paper explicitly distinguishes its unconstrained reweighting from KMM rather than renaming it. The main weakness, that aligning to a stale EWMA may not reduce true gradient error under drift, is an unproven correctness/robustness concern, not a circularity. Therefore no circular step meeting the required standard is present.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central mechanism rests on the EWMA being a proxy for the true discrepancy and on the chain-rule proportionality between gradient and statistic errors; both are domain assumptions, not derived. alpha and the buffer threshold are hand-set, and lambda/k are tuned per sampler. No new physical entities are introduced.

free parameters (5)
  • alpha (EWMA decay) = 0.1
    Set for all experiments; controls how quickly the reference statistic tracks the stream and therefore the weighting objective.
  • buffer coefficient cutoff = 0.01
    Minibatches with EWMA coefficients below 0.01 are discarded, capping memory at ~22 minibatches; this truncates the reference statistic.
  • lambda (UDA trade-off) = tuned per sampler via random search
    Balances task loss and discrepancy loss in Eq. (1); tuned independently for each sampler, which can mask algorithmic differences.
  • minibatch size k = sampled from 2^Uniform(3,7)
    Part of the per-sampler random search; affects estimator variance and runtime.
  • RBF kernel bandwidth set G = {0.001, 0.01, 0.1, 1, 10}
    Kernel mixture bandwidths fixed as in prior work; chosen by hand, not fitted to target accuracy, but still a modeling choice.
axioms (5)
  • standard math The kernel is positive-definite, so the MMD weight-optimization objective is a convex quadratic.
    Used in Section 2.2 Eq. (15) to justify efficient solution; requires the RKHS/kernel setup.
  • domain assumption The discrepancy gradient error is proportional to the statistic error (D_hat - D), so minimizing ||D_hat_w - D_tilde|| reduces gradient variance under smoothness conditions.
    Stated in Section 2.1, citing Zhao & Zhang (2014) and Liu et al. (2020); no proof is provided in this paper.
  • domain assumption The EWMA D_tilde_m is an adequate stand-in for the true time-varying discrepancy D_m.
    The method targets ||D_hat_w - D_tilde_m|| rather than ||D_hat_w - D_m||; if the EWMA is biased or lags, the variance-reduction claim fails.
  • ad hoc to paper The CORAL relaxation (linear combination of outer products with fixed centers) preserves enough of the covariance signal to reduce variance.
    Section 2.3 introduces this relaxation to avoid an NP-hard quartic problem; no justification is given that the relaxation retains the variance-reduction property.
  • ad hoc to paper Unconstrained weights can be optimized to align statistics without exploding gradient norms.
    Section 2.2 explicitly drops probability constraints; with no regularization, weights can grow large, but the paper does not analyze this.

pith-pipeline@v1.3.0-alltime-deepseek · 12999 in / 14196 out tokens · 115820 ms · 2026-08-01T09:59:43.262321+00:00 · methodology

0 comments
read the original abstract

This paper studies the problem of stochastic variance reduction (SVR) for the maximum mean discrepancy (MMD) and correlation alignment (CORAL) loss functions. Although various offline SVR algorithms for these losses have been proposed, these are incompatible with online, distributed, or incremental learning settings. This paper presents Adaptive vaRiance Reduction via Online reWeighting (ARROW), the first online SVR algorithm for the MMD and CORAL for streamed data. The method maintains moving average references of the alignment statistics, and adaptively reweights incoming minibatches so that the minibatch and reference statistics are aligned. Further, we propose a relaxed reweighting scheme so that the ensuing weight-optimisation problem is tractable. In experiments and simulations, we show that ARROW performs competitively with offline algorithms in terms of runtime, degree of variance reduction achieved, and target domain accuracy.

Figures

Figures reproduced from arXiv: 2607.20374 by Andrea Napoli.

Figure 1
Figure 1. Figure 1: Estimator variance vs k for six sampling algorithms. with fixed (unweighted) centres bc m s , bc m t . Thus, we obtain a tractable linear least squares problem min u,v [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

27 extracted references · 16 linked inside Pith

  1. [6]

    doi: 10.1109/CVPR46437.2021.01411

    ISSN 10636919. doi: 10.1109/CVPR46437.2021.01411. URLhttps: //arxiv.org/abs/2103.15796v2. 8 Tianfan Fu and Zhihua Zhang. CPSG-MCMC: Clustering-Based Preprocessing method for Stochastic Gradient MCMC.AISTATS, pp. 841–850, 4

    arXiv 2021

  2. [7]

    Deep Residual Learning for Image Recognition.Pro- ceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2016-December: 770–778, 12

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep Residual Learning for Image Recognition.Pro- ceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2016-December: 770–778, 12

    2016

  3. [10]

    URLhttps://link.springer.com/article/10.1007/s10994-009-5152-4

    1007/S10994-009-5152-4/METRICS. URLhttps://link.springer.com/article/10.1007/s10994-009-5152-4. Yoshua Bengio, Jérome Louradour, Ronan Collobert, and Jason Weston. Curriculum learning.ICML, 382,

    doi:10.1007/s10994-009-5152-4

  4. [12]

    Weijie Liu, Hui Qian, Chao Zhang, Zebang Shen, Jiahao Xie, and Nenggan Zheng

    URLhttps://arxiv.org/abs/1810.03124v1. Weijie Liu, Hui Qian, Chao Zhang, Zebang Shen, Jiahao Xie, and Nenggan Zheng. Accelerating Stratified Sampling SGD by Reconstructing Strata.IJCAI,

    Pith/arXiv arXiv

  5. [13]

    URL https://proceedings.mlr.press/v37/long15.html

    PMLR. URL https://proceedings.mlr.press/v37/long15.html. Mingsheng Long, Zhangjie Cao, Jianmin Wang, and Michael I. Jordan. Conditional Adversarial Domain Adaptation. Advances in Neural Information Processing Systems, 2018-December:1640–1650, 5

    2018

  6. [15]

    URL https://arxiv.org/abs/2103.02062v2

    ISSN 26403498. URL https://arxiv.org/abs/2103.02062v2. Aengus Lynch, Gbètondji J-S Dovonon, Jean Kaddour, and Ricardo Silva. Spawrious: A Benchmark for Fine Control of Spurious Correlation Biases.arXiv, 3

    Pith/arXiv arXiv

  7. [17]

    Diversity-Based Sampling for Imbalanced Domain Adaptation.EUSIPCO, 2024a

    Andrea Napoli and Paul White. Diversity-Based Sampling for Imbalanced Domain Adaptation.EUSIPCO, 2024a. 9 Andrea Napoli and Paul White. Improving Domain Generalisation with Diversity-based Sampling.DCASE, 2024b. URLhttp://arxiv.org/abs/2410.04235. Andrea Napoli and Paul White. Order Matters: Improving Domain Adaptation by Reordering Data.arXiv, 2026a. And...

    Pith/arXiv arXiv

  8. [18]

    doi: 10.1037/met0000301

    ISSN 1082989X. doi: 10.1037/met0000301. URLhttps://pubmed.ncbi.nlm.nih. gov/32567870/. Harsh Rangwani, Sumukh K Aithal, Mayank Mishra, Arihant Jain, and R Venkatesh Babu. A Closer Look at Smooth- ness in Domain Adversarial Training.Proceedings of the 39th International Conference on Machine Learning,

    arXiv

  9. [20]

    Vladimir Vapnik.Statistical Learning Theory

    URLhttps://arxiv.org/abs/1412.3474v1. Vladimir Vapnik.Statistical Learning Theory. Wiley, New York, US,

    Pith/arXiv arXiv

  10. [22]

    URLhttps://arxiv

    doi: 10.1109/CVPR.2017.572. URLhttps://arxiv. org/abs/1706.07522v1. Zirui Wang, Zihang Dai, Barnabas Poczos, and Jaime Carbonell. Characterizing and Avoiding Negative Transfer. CVPR, 2019-June:11285–11294, 11

    Pith/arXiv arXiv 2017

  11. [24]

    doi: 10.1609/AAAI.V33I01.33015741

    ISSN 2374-3468. doi: 10.1609/AAAI.V33I01.33015741. URL https://ojs.aaai.org/index.php/AAAI/article/view/4520. Marvin Zhang, Henrik Marklund, Nikita Dhawan, Abhishek Gupta, Sergey Levine, and Chelsea Finn. Adaptive Risk Minimization: Learning to Adapt to Domain Shift.Advances in Neural Information Processing Systems, 28: 23664–23678, 7

    doi:10.1609/aaai.v33i01.33015741

  12. [25]

    URLhttps://arxiv.org/abs/2007.02931v4

    ISSN 10495258. URLhttps://arxiv.org/abs/2007.02931v4. YiFan Zhang, Xue Wang, Jian Liang, Zhang Zhang, Liang Wang, Rong Jin, and Tieniu Tan. Free Lunch for Domain Adversarial Training: Environment Label Smoothing.ICLR, 2

    Pith/arXiv arXiv 2007

  13. [26]

    Peilin Zhao and Tong Zhang

    URLhttp://arxiv.org/abs/2302.00194. Peilin Zhao and Tong Zhang. Accelerating Minibatch Stochastic Gradient Descent using Stratified Sampling.arXiv, 5

    Pith/arXiv arXiv

  14. [27]

    URLhttps://arxiv.org/abs/1405.3080v1. 10

    Pith/arXiv arXiv

  15. [1998]

    Deep Hashing Network for Unsupervised Domain Adaptation.CVPR 2017,

    Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep Hashing Network for Unsupervised Domain Adaptation.CVPR 2017,

    2017

  16. [2007]

    7 Table 3: Average test accuracy for Humpbacks by data split. Method Domain 1 Domain 2 Domain 3 Domain 4 A verage ERM 70.3±2.7 92.0±1.9 78.1±3.0 96.2±0.6 84.2±1.1 DANN 60.5±4.2 90.1±2.1 63.9±6.1 76.1±11.1 72.6±3.4 CDAN 61.6±4.2 82.2±4.7 73.5±3.0 84.4±0.6 75.4±1.7 CDAN + SDAT 63.7±3.0 81.2±6.7 63.6±3.7 78.9±3.2 71.8±2.2 CDAN + ELS 62.7±2.4 85.6±3.4 70.8±2....

    1944

  17. [2009]

    URLhttps://dl.acm.org/doi/10.1145/1553374.1553380

    doi: 10.1145/1553374.1553380. URLhttps://dl.acm.org/doi/10.1145/1553374.1553380. Abhimanyu Dubey, Vignesh Ramanathan, Alex Pentland, and Dhruv Mahajan. Adaptive Methods for Real-World Domain Generalization.CVPR,

    arXiv

  18. [2014]

    URLhttps://arxiv.org/abs/1412.6980v9

    doi: 10.48550/arxiv.1412.6980. URLhttps://arxiv.org/abs/1412.6980v9. Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, Tony Lee, Etienne David, Ian Stavness, Wei Guo, Berton A. Earnshaw, Imran S. Haque, Sara Beery, Jure Leskovec, Anshul Kunda...

    Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980

  19. [2015]

    doi: 10.48550/arxiv.1512.03385

    ISSN 10636919. doi: 10.48550/arxiv.1512.03385. URLhttps://arxiv.org/abs/1512.03385v1. Wenyu Jiang, Hao Cheng, MingCai Chen, Chongjun Wang, and Hongxin Wei. DOS: Diverse Outlier Sampling for Out-of-Distribution Detection.ICLR,

    Pith/arXiv arXiv doi:10.48550/arxiv.1512.03385

  20. [2016]

    URLhttps://arxiv.org/abs/1607.01719v1

    ISSN 16113349. URLhttps://arxiv.org/abs/1607.01719v1. Eric Tzeng, Judy Hoffman, Ning Zhang, Kate Saenko, and Trevor Darrell. Deep Domain Confusion: Maximizing for Domain Invariance.arXiv, 12

    Pith/arXiv arXiv

  21. [2017]

    URL https://arxiv.org/abs/1705.10667v4

    ISSN 10495258. URL https://arxiv.org/abs/1705.10667v4. Yucheng Lu, Youngsuk Park, Lifan Chen, Yuyang Wang, Christopher De Sa, and Dean Foster. Variance Reduced Training with Stratified Sampling for Forecasting Models.ICML, 139:7145–7155, 3

    Pith/arXiv arXiv

  22. [2018]

    doi: 10.1109/CVPR.2018.00566

    ISSN 10636919. doi: 10.1109/CVPR.2018.00566. Jingchang Liu and Linli Xu. Accelerating Stochastic Gradient Descent Using Antithetic Sampling.arXiv, 10

    arXiv 2018

  23. [2019]

    doi: 10.1109/CVPR.2019.01155

    ISSN 10636919. doi: 10.1109/CVPR.2019.01155. URLhttps://arxiv. org/abs/1811.09751v4. Lilei Wu and Jie Liu. Contrastive Learning with Diverse Samples.Frontiers in Artificial Intelligence and Applications, 372:2672–2679, 9

    arXiv 2019

  24. [2020]

    doi: 10.1007/978-3-030-58589-1{\_}28

    ISSN 16113349. doi: 10.1007/978-3-030-58589-1{\_}28. URL https://arxiv.org/pdf/1912.03699. Diederik P. Kingma and Jimmy Lei Ba. Adam: A Method for Stochastic Optimization.3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings, 12

    Pith/arXiv arXiv 1912

  25. [2021]

    URL https://arxiv.org/abs/2112.06007v1

    ISSN 10495258. URL https://arxiv.org/abs/2112.06007v1. Philipp Baumann, Olivier Goldschmidt, Dorit S. Hochbaum, and Jason Yang. A Fast and Effective Method for Euclidean Anticlustering: The Assignment-Based-Anticlustering Algorithm.arXiv, 1

    Pith/arXiv arXiv

  26. [2023]

    Andrea Napoli

    URLhttps://arxiv.org/abs/2303.05470v3. Andrea Napoli. Variance-reduced Domain Adaptation using Paired Sampling.arXiv,

    Pith/arXiv arXiv

  27. [2026]

    org/abs/2601.06351

    URLhttp://arxiv. org/abs/2601.06351. Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of Representations for Domain Adaptation.NeurIPS, 19,

    Pith/arXiv arXiv