Pith. sign in

REVIEW 6 major objections 5 minor 34 references

This paper claims that actively controlling the mini-batch sampling distribution with frequency and sharpness feedback—not post-hoc perturbation correction—moves long-tailed training toward balanced flat minima, and it backs that claim with

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 06:05 UTC pith:KEDTEL7J

load-bearing objection A genuinely new sampling-side idea with large reported tail-accuracy gains, but the theory is conditional to the point of circularity and the evaluation has protocol asymmetries that need fixing before the numbers can be trusted. the 6 major comments →

arxiv 2607.21999 v1 pith:KEDTEL7J submitted 2026-07-24 cs.LG cs.AI

From Perturbation Correction to Geometry-Aware Sampling: Sharpness-Guided Equilibrium Sampling for Balanced Flat Minima in Long-Tailed Learning

Jiaxin Deng , Junbiao Pang This is my paper
classification cs.LG cs.AI
keywords long-tailed learningsharpness-aware minimizationgeometry-aware samplingflat minimaclass imbalancesampling distribution controlPAC-Bayes generalization boundCIFAR-LT / ImageNet-LT
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.

The paper tries to establish that in long-tailed learning, the sampling distribution should be treated as an active control variable for optimization geometry rather than a fixed prior that later methods compensate for. It introduces Sharpness-Guided Equilibrium Sampling (SGS), which raises the sampling probability of classes that have been sampled less often while lowering it for classes whose SAM-induced loss changes are large, using only cumulative class counts and EMA sharpness estimates already produced by the standard SAM update. The paper characterizes this process with a stochastic differential equation whose stationary density concentrates on low frequency-curvature classes, and with a sampling-dependent PAC-Bayes bound claimed to be tighter than ordinary long-tailed SAM. Empirically, SGS-SAM improves tail accuracy by 10.85 points over Focal-SAM on CIFAR-100 LT (IR100) and by 6.59 points over ImbSAM on ImageNet-LT, at 1.02× vanilla SAM's training cost. A sympathetic reader would care because it suggests a sampling-side, nearly free route to controlling loss-landscape geometry, complementing perturbation- and loss-side methods.

Core claim

The central claim is that the mini-batch sampler itself can be the mechanism that steers optimization toward balanced flat minima. SGS maintains per-class cumulative counts m_c(t) and EMA sharpness h_c(t), where h_c is an exponential moving average of the per-example loss gap |ℓ_i(w+ε)−ℓ_i(w)| from the standard SAM step. Each class's sampling weight is w_c = [1/(p_c+ξ)] · [1/(h_c^κ+ξ)], so a class receives more samples if it has been under-sampled and fewer if it is currently sharp; this weight is combined with the base prior to form the next mini-batches. The authors prove, under stated conditions, that the continuous-time limit of this closed loop has a stationary density that suppresses o

What carries the argument

The load-bearing object is the class-level sampling weight w_c = w_freq_c · w_sharp_c = 1/(p_c+ξ) · 1/(h_c^κ+ξ), where p_c(t) is the cumulative sampling share of class c and h_c(t) is the EMA of its SAM-induced loss gaps. This product couples frequency and geometry: the first factor pulls back under-sampled classes, the second pushes away classes in sharp regions. The SAM loss gap ΔL_x = |ℓ_x(w+ε) − ℓ_x(w)| acts as a zero-cost sharpness proxy because the quadratic term in its Taylor expansion captures directional curvature. Theorems 1 and 2 convert this discrete rule into an SDE whose stationary density φ(q) ∝ q^{−γ}, γ = 1 + 2(μ_c − μ̄)/σ̃², providing the 'equilibrium' for which the method

Load-bearing premise

The PAC-Bayes theorem assumes what it needs to prove: that the stationary SGS distribution q* is closer to the balanced prior than the empirical long-tailed prior (Eq. 14) and that the increase in sharp empirical risk is outweighed by the decrease in the PAC penalty; the paper does not derive these inequalities from the SGS update rule, so the theoretical explanation could fail even if the empirical gains hold.

What would settle it

Train SGS with the sharpness feedback inverted (weight w_c ∝ w_freq · h_c^κ, so sharp classes are up-sampled) on CIFAR-100 LT IR100 under identical hyperparameters; if tail accuracy remains near 19.75, the sharpness term is not the cause of the gains. A second check: record the actual final class sampling shares p_c(T) under SGS and test whether ∥q* − u∥₁ < ∥π − u∥₁; if this inequality fails, the premise of Theorem 3 is violated.

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

If this is right

  • SGS changes where optimization happens, not just the loss: sampling probability of under-exposed, low-curvature classes increases automatically over time.
  • Because it reuses the SAM perturbation output, SGS adds no backward passes or class-wise perturbations; measured cost is about 1.02× vanilla SAM.
  • It is compatible with deferred logit adjustment: representation learning uses SGS-SAM, then classifier calibration with LA, yielding additional tail-accuracy gains.
  • The SDE view predicts an automatic 'thermostat': if a tail class becomes sharp, its sampling weight drops, preventing overfitting to unstable regions.
  • Under severe imbalance (IR100/200, ImageNet-LT) the gains concentrate in medium and tail classes; under mild imbalance (IR10) SGS is comparable to Focal-SAM, so its value scales with skew.

Where Pith is reading between the lines

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

  • The principle is general: any optimizer that yields a per-example loss gap (e.g., adversarial training, noise injection) could drive a similar sharpness-guided sampler without SAM, so SGS could be lifted to other domains the paper does not explore.
  • The sharpness EMA h_c acts like an adaptive temperature; annealing κ over training might yield a curriculum that first rebalances exposure and then enforces flatness, a natural extension not studied here.
  • The class-level weighting assigns equal instance probability within a class; an instance-level variance term could further distinguish easy versus hard tail examples, a testable modification.
  • If the reported gap between κ=0.6 and small κ is real, the sharpness term is doing substantial work; quantifying the difference between SGS and pure frequency rebalancing under identical compute is the cleanest causal test.

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

6 major / 5 minor

Summary. This paper proposes Sharpness-Guided Equilibrium Sampling (SGS), a class-level adaptive sampling strategy for long-tailed classification. SGS maintains cumulative class counts and per-class EMA estimates of the SAM perturbation loss gap, and uses the inverse product of these quantities to reweight the class sampling distribution after each mini-batch. The authors provide a continuous-time SDE analysis (Theorems 1–2) and a PAC-Bayes argument (Theorem 3) intended to show that SGS moves training toward a balanced flat minimum. They report large gains on CIFAR-10/100-LT, ImageNet-LT, and CLIP fine-tuning, and claim a training time of 1.02× vanilla SAM.

Significance. The core idea is attractive and potentially useful: reusing the SAM perturbation response as a cheap class-level sharpness signal to steer mini-batch construction is novel, and the method is lightweight, requiring no class-wise perturbations or extra backward passes. The reported tail-class improvements are substantial and the extension to foundation-model fine-tuning is a useful stress test. At the same time, the theoretical support is conditional in a way that does not establish the claimed mechanism, and the empirical comparison contains several protocol mismatches. In particular, the paper never isolates the sharpness term from plain frequency rebalancing, and the 1.02× speed figure is not measured. If these issues are fixed, the contribution could be valuable; as it stands, the evidence is suggestive rather than conclusive.

major comments (6)
  1. [A Stochastic Dynamics Perspective of SGS, Theorem 3 (Eqs. 13–15)] Theorem 3 assumes essentially the desired conclusion. Eq. (14), namely ∥q*−u∥₁ < ∥π−u∥₁, and the condition that any increase in sharp empirical risk is smaller than the reduction in B are imposed as suppositions; they are never derived from the SGS update rule (Eqs. 1–9) or from Theorem 2. The 'equalizing-transfer decomposition' is also introduced without a definition. Thus the theorem does not explain why SGS should yield a tighter PAC-Bayes bound; it merely states that if the bound improves, then the bound improves. Please either derive these conditions from the update rule under explicit, verifiable assumptions, or clearly label the argument as a conjecture/heuristic and soften the corresponding claims in the abstract and introduction.
  2. [Theorems 1 and 2] Both theorems rest on unproved continuous-time assumptions: 'Assume g_c(t) admits the Stratonovich limit' and 'Assume the remaining class-wise noise can be represented by an effective volatility σ~_c with reflecting boundaries on [q_min, q_max]'. No discrete-to-continuum derivation, scaling argument, or empirical consistency check links the actual mini-batch process to the SDE. As written, the SDE analysis is a formal analogy rather than a theorem about SGS. Please provide a rigorous limit or explicitly state that this is a heuristic analogy and remove the claim that it constitutes a proof of SGS's behavior.
  3. [Long-Tailed Classification, 'Comparison on CIFAR10/100-LT' (Table 3)] The LA+SGS-SAM results use a deferred LA protocol: the model is first trained with SGS-SAM for 180 epochs, then the backbone is frozen and only the classifier is fine-tuned for 20 epochs with LA+SAM. The LA baselines (LA, LA+SAM, LA+ImbSAM, LA+CC-SAM, LA+Focal-SAM) do not receive this deferred classifier fine-tuning. This protocol mismatch confounds the comparison; the large tail gains under LA could be due to the extra classifier fine-tuning stage rather than to SGS. Please apply the same deferred protocol to all LA baselines, or report SGS without deferred fine-tuning, or use a standard matched protocol for all methods.
  4. [Large-Scale Long-Tailed Fine-Tuning (Table 5)] SGS-SAM25 and LIFT+SGS-SAM25 are trained for 25 epochs, while all compared methods and the default protocol use 20 epochs. The 25-epoch results are therefore not directly comparable to the 20-epoch baselines. The claim that SGS 'continues to improve beyond the default schedule' does not justify comparing a 25-epoch method against 20-epoch results. Please report 20-epoch SGS results under the same protocol, or compare against baselines also trained for 25 epochs.
  5. [Method and Experiments (Eq. (5), Table 3)] No ablation disables the sharpness term (κ=0) or compares SGS against a frequency-only sampler. Since SGS combines inverse cumulative frequency and inverse EMA sharpness, the empirical gains could be entirely due to frequency rebalancing. Add a frequency-only baseline (e.g., sampling proportionally to 1/p_c(t) without h_c) and a κ=0 ablation on at least CIFAR-100-LT IR100. This is necessary to support the 'sharpness-guided' claim and to distinguish the contribution from existing re-sampling approaches.
  6. [Training Speed (Table 6)] The text states that the comparison ratios in Table 6 'are taken from the Focal-SAM (Li et al. 2025)' paper, not measured here. This means the 1.02× figure for SGS-SAM is not an empirical measurement from this work, and the abstract's speed claim is unsupported. Please measure wall-clock time on the same hardware with the same implementation, or remove/reword the speed claim and clearly state that the figure is an estimate from prior work.
minor comments (5)
  1. [Throughout] The paper alternates between 'SGS' and 'SGES' (e.g., in Related Works and the Conclusion). Please use one consistent acronym.
  2. [References to Appendix] The manuscript repeatedly references an appendix ('More details about datasets, networks, and training settings are presented in Appendix' and 'the reasons why SGS needs more epochs for better results is discussed in Appendix'), but no appendix is included in the provided text. Please include the appendix or remove the references.
  3. [Theorem 3, Eq. (13)] The objects u, π, and q* are not formalized. Define the balanced prior u, specify the PAC complexity terms Φ(q_c), and state the assumed loss bounds explicitly.
  4. [Tables 1 and 3] Table 1 reports the best IR100 result at r_w=0.7 as 47.06 overall and 20.76 tail, while Table 3 reports CE+SGS-SAM as 47.56 overall and 19.75 tail. Clarify whether these are different runs/configurations or whether the tables are inconsistent.
  5. [Table 5 and Table 6] The text states that SGS-SAM25's total training time is 1.26× vanilla SAM, but this figure is not included in Table 6. If speed is claimed, report it consistently in the speed table.

Circularity Check

1 steps flagged

Theorem 3's tighter-bound conclusion is an algebraic restatement of its own risk-increase assumption; Eq. (14) is assumed, not derived from SGS.

specific steps
  1. self definitional [Theorem 3, Eqs. (14)-(15)]
    "Suppose that the movement from π to q⋆ admits an equalizing-transfer decomposition in the effective-sample region, that q⋆ is closer to the balanced prior, ∥q⋆−u∥₁ < ∥π−u∥₁, (14) and that any increase in sharp empirical risk is smaller than the resulting reduction in B. Then, with probability at least 1−δ, SGS has a tighter PAC-Bayes right-hand side than ordinary long-tailed SAM"

    The conclusion (15), bLρ_{q⋆,S}(w)+B(q⋆) < bLρ_{π,S}(w)+B(π), is exactly the rearrangement of the hypothesis that any increase in sharp empirical risk is smaller than the reduction in B: bLρ_{q⋆,S}(w)-bLρ_{π,S}(w) < B(π)-B(q⋆). Thus the theorem's result is its own antecedent restated. The additional condition (14), ∥q⋆−u∥₁ < ∥π−u∥₁, is never derived from the SGS update rules (Eqs. 1–9) or from Theorems 1–2. Theorem 2 only gives the stationary density form q^{-γ} after freezing coefficients and imposing boundary conditions; the remark after Theorem 3 asserts 'As a result, ∥q⋆−u∥1 is reduced' without proof. Hence the PAC-Bayes argument does not show SGS moves training toward balanced flat minima; it assumes that movement.

full rationale

The empirical comparisons are independent and not circular: SGS-SAM is evaluated on held-out CIFAR-LT/ImageNet-LT test splits against published baselines, and the reported accuracy gains are not fitted renames of the method's inputs. There are no load-bearing self-citations. However, the theoretical narrative is partially circular: Theorem 3's tighter-bound result reduces by construction to the assumption that the sharp-risk increase is smaller than the reduction in B (Eq. 15 is the rearrangement of that condition), and Eq. (14), stating that q⋆ is closer to the balanced prior, is assumed rather than derived from the SGS update or Theorems 1–2. Consequently, the paper's explanation of why frequency-sharpness feedback yields a balanced flat-minima profile is conditional on the very effect it claims to prove. The absence of a κ=0 ablation further weakens the empirical support for the sharpness-specific mechanism, but that is an experimental gap rather than circularity. Score 6 reflects partial circularity in the theoretical claim while the empirical results retain independent content.

Axiom & Free-Parameter Ledger

5 free parameters · 7 axioms · 0 invented entities

The method has several tuned hyperparameters (κ, r_w, β, α_max, ξ) and a free mixing schedule. The theoretical results rest on assumptions that include the desired conclusion, most notably Theorem 3's Eq. (14) and the risk-increase condition. No new physical entities are introduced.

free parameters (5)
  • β (EMA momentum) = 0.9 or 0.99
    Chosen by hand in Eq. (4); controls stability of class sharpness estimates; no principled selection.
  • κ (sharpness exponent) = 0.6 for CIFAR-100 IR100; 0.9 gives best tail; 2.0 for fine-tuning
    Table 2 shows tail accuracy varies from 13.90 to 23.62 across κ; κ=0.6 maximizes overall accuracy and is adopted for final results.
  • r_w (warm-up ratio) = 0.7 for CIFAR-100 IR100; 0.4 for IR200
    Table 1 shows tail accuracy varies from 12.83 to 20.76 with r_w; different imbalance ratios require different warm-ups.
  • α_max and α(e) mixing schedule = 1.0 for CIFAR/ImageNet; 0.1 for fine-tuning; schedule unspecified
    Eq. (9) defines the split between base and SGS sampling, but the schedule is only described as 'gradually increases'; the fraction of SGS samples directly controls the method's strength.
  • ξ (numerical stability constant) = unspecified
    Used in Eqs. (1), (5), and (7) to avoid division by zero; value not reported.
axioms (7)
  • domain assumption The per-sample SAM loss gap |L_xi(w+ε) − L_xi(w)| is a valid class-level sharpness proxy via second-order Taylor expansion.
    Invoked in Eq. (2) and surrounding text; the paper does not compare this proxy to Hessian-based or trajectory-based flatness measures, and for tail classes the EMA is updated only when the class appears in a batch.
  • ad hoc to paper The class score g_c(t) admits a Stratonovich stochastic differential equation dg_c = μ_c dt + σ_c ∘ dW_c.
    Theorem 1 simply assumes this limit exists; no derivation from the discrete update rule in Algorithm 1 is provided.
  • ad hoc to paper The movement from π to q⋆ admits an equalizing-transfer decomposition and satisfies ∥q⋆−u∥₁ < ∥π−u∥₁.
    Assumed in Theorem 3 (Eq. 14); the paper does not show that the SGS update rule produces such a q⋆.
  • ad hoc to paper Any increase in sharp empirical risk under q⋆ is smaller than the reduction in the PAC-Bayes penalty B(q⋆).
    Assumed in Theorem 3 immediately before Eq. (15); this is the conclusion restated as a premise.
  • standard math PAC-Bayes bounds for SAM (Foret et al. 2021) apply under class-level adaptive sampling with effective sample size n q_c.
    Standard PAC-Bayes machinery, but the extension to adaptive, non-i.i.d. sampling is not proved.
  • domain assumption Class-balanced accuracy (prior u) is the correct evaluation target and the balanced-prior mismatch ∥q−u∥₁ is the right complexity penalty.
    Postulated in Theorem 3; common in long-tailed literature but still a modeling choice.
  • ad hoc to paper In Theorem 2, coefficients are frozen and an effective scalar volatility σ̃_c with reflecting boundaries on [q_min, q_max] approximates the full class-wise noise.
    Local approximation; no error bound connects this stationary density to the actual discrete process.

pith-pipeline@v1.3.0-alltime-deepseek · 12814 in / 18279 out tokens · 173847 ms · 2026-08-01T06:05:56.600290+00:00 · methodology

0 comments
read the original abstract

Long-tailed learning couples two sources of poor generalization: head classes dominate training exposure, while under-represented classes often converge to sharper regions of the loss landscape. Conventional re-sampling addresses the former without considering geometry, whereas existing long-tailed sharpness-aware minimization (SAM) methods modify losses or perturbations only after biased mini-batches have been drawn. We introduce Sharpness-Guided Equilibrium Sampling (SGS), which treats the sampling distribution as an active control variable for optimization geometry. SGS dynamically adjusts subsequent mini-batches by increasing the sampling probability of less frequently sampled classes while suppressing classes with large SAM-induced loss changes, using only cumulative class counts and EMA sharpness estimates obtained from the standard SAM update, without class-wise perturbations or additional backward passes. We characterize this sampling process through a continuous-time stochastic differential equation and a sampling-dependent PAC-Bayes analysis, explaining how frequency-sharpness feedback can move training toward a more balanced flatness profile. On CIFAR-100 LT with an imbalance ratio of 100, SGS-SAM improves Focal-SAM by 10.85 points in tail accuracy and 3.56 points overall. On ImageNet-LT, it improves ImbSAM by 6.59 points on tail classes and 1.20 points overall. Its training time is only $1.02\times$ that of vanilla SAM. Beyond these gains, SGS establishes a sampling-side route to loss-landscape control, suggesting that future long-tailed methods can jointly regulate data exposure and optimization geometry rather than treating either as fixed.

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

34 extracted references · 3 linked inside Pith

  1. [1]

    2020 IEEE international conference on big data (Big data) , pages=

    Pyhessian: Neural networks through the lens of the hessian , author=. 2020 IEEE international conference on big data (Big data) , pages=. 2020 , organization=

    2020

  2. [2]

    Journal of Statistical Mechanics: Theory and Experiment , volume=

    Entropy-sgd: Biasing gradient descent into wide valleys , author=. Journal of Statistical Mechanics: Theory and Experiment , volume=. 2019 , publisher=

    2019

  3. [3]

    Proceedings of the International Conference on Learning Representations (ICLR) , year=

    On large-batch training for deep learning: Generalization gap and sharp minima , author=. Proceedings of the International Conference on Learning Representations (ICLR) , year=

  4. [4]

    arXiv preprint arXiv:1711.04623 , year=

    Three factors influencing minima in sgd , author=. arXiv preprint arXiv:1711.04623 , year=

    Pith/arXiv arXiv

  5. [5]

    European conference on computer vision , pages=

    Microsoft coco: Common objects in context , author=. European conference on computer vision , pages=. 2014 , organization=

    2014

  6. [6]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Lvis: A dataset for large vocabulary instance segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  7. [7]

    International Conference on Learning Representations , year=

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , author=. International Conference on Learning Representations , year=

  8. [8]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  9. [9]

    International Conference on Learning Representations , year=

    Long-tail learning via logit adjustment , author=. International Conference on Learning Representations , year=

  10. [10]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  11. [11]

    Learning multiple layers of features from tiny images.(2009) , author=

    2009

  12. [12]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    The inaturalist species classification and detection dataset , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  13. [13]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Large-scale long-tailed recognition in an open world , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  14. [14]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Internimage: Exploring large-scale vision foundation models with deformable convolutions , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  15. [15]

    Advances in neural information processing systems , volume=

    Yolov10: Real-time end-to-end object detection , author=. Advances in neural information processing systems , volume=

  16. [16]

    International Conference on Machine Learning , pages=

    Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification , author=. International Conference on Machine Learning , pages=. 2025 , organization=

    2025

  17. [17]

    Neural networks , volume=

    A systematic study of the class imbalance problem in convolutional neural networks , author=. Neural networks , volume=. 2018 , publisher=

    2018

  18. [18]

    Advances in Neural Information Processing Systems , volume=

    A unified generalization analysis of re-weighting and logit-adjustment for imbalanced learning , author=. Advances in Neural Information Processing Systems , volume=

  19. [19]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Generalized parametric contrastive learning , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2023 , publisher=

    2023

  20. [20]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    Long-tail learning with foundation model: heavy fine-tuning hurts , author=. Proceedings of the 41st International Conference on Machine Learning , pages=

  21. [21]

    Advances in neural information processing systems , volume=

    Learning imbalanced datasets with label-distribution-aware margin loss , author=. Advances in neural information processing systems , volume=

  22. [22]

    International Conference on Learning Representations , year=

    Sharpness-aware Minimization for Efficiently Improving Generalization , author=. International Conference on Learning Representations , year=

  23. [23]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Imbsam: A closer look at sharpness-aware minimization in class-imbalanced recognition , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  24. [24]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Class-conditional sharpness-aware minimization for deep long-tailed recognition , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  25. [25]

    arXiv preprint arXiv:1910.09217 , year=

    Decoupling representation and classifier for long-tailed recognition , author=. arXiv preprint arXiv:1910.09217 , year=

    Pith/arXiv arXiv 1910

  26. [26]

    Advances in neural information processing systems , volume=

    Balanced meta-softmax for long-tailed visual recognition , author=. Advances in neural information processing systems , volume=

  27. [27]

    European conference on computer vision , pages=

    The devil is in classification: A simple framework for long-tail instance segmentation , author=. European conference on computer vision , pages=. 2020 , organization=

    2020

  28. [28]

    arXiv preprint arXiv:2007.07314 , year=

    Long-tail learning via logit adjustment , author=. arXiv preprint arXiv:2007.07314 , year=

    Pith/arXiv arXiv 2007

  29. [29]

    International Conference on Machine Learning , pages=

    Feature directions matter: Long-tailed learning via rotated balanced representation , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  30. [30]

    The twelfth international conference on learning representations , year=

    Long-tailed diffusion models with oriented calibration , author=. The twelfth international conference on learning representations , year=

  31. [31]

    Science China Information Sciences , volume=

    Prototype-based classifier learning for long-tailed visual recognition , author=. Science China Information Sciences , volume=. 2022 , publisher=

    2022

  32. [32]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Parametric contrastive learning , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  33. [33]

    European conference on computer vision , pages=

    Safa: Sample-adaptive feature augmentation for long-tailed image classification , author=. European conference on computer vision , pages=. 2022 , organization=

    2022

  34. [34]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Improving calibration for long-tailed recognition , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=