Pith. sign in

REVIEW 4 major objections 6 minor 104 references

Generative Data Mining with Longtail-Guided Diffusion

T0 review · 4 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read By decoding noisy latent states back to images and steering the diffusion update with the production model's longtail gradient, Longtail Guidance generates rare or hard training data without retraining either model.

desk verdict A practical Epistemic Head and a clever terminal-state guidance trick, but the benchmark gains are confounded by iterative generation scheduling; worth reviewing with a controlled comparison. read the letter →

arxiv 2502.01980 v2 pith:KDF5MD67 submitted 2025-02-04 cs.LG cs.AI

David S. Hayden , Mao Ye , Timur Garipov , Gregory P. Meyer , Carl Vondrick , Zhao Chen , Yuning Chai , Eric Wolff
show 1 more author
Siddhartha S. Srinivasa
This is my paper · ORCID
classification cs.LGcs.AI
keywords longtailguidancesyntheticdatagenerationepistemicuncertaintylatentdiffusionclassifierimageclassificationlong-tailrecognitiongenerativemining
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

The paper proposes Longtail Guidance (LTG), a way to turn an existing image classifier's difficulties into a steering signal for a latent diffusion model, generating new training images that are rare or hard for that classifier. The claim is that by differentiating through a decoded estimate of the clean image, the production model does not need to see noisy diffusion states and can be used off-the-shelf. Fine-tuning the classifier on LTG-generated data is reported to improve top-1 accuracy by an average of 5.6 points over GIF and Dream-ID and 15.0 points over prompt-tuned Stable Diffusion, with the largest gains on the rarest ImageNet-LT classes. The paper also turns LTG-generated images into text keywords via a vision-language model, so the discovered longtail can be inspected and prompted for.

What carries the argument

The load-bearing object is the decoded terminal-state estimate $\hat{x}^{(t)}_0 = D(P(z_t))$: at every DDIM step, the latent state $z_t$ is mapped to its predicted clean latent and decoded through the VAE, then scored by the production model. Because this estimate stays close to real training data much earlier in the reverse process, which the paper measures with FID, the production model's gradient is meaningful without being trained on noisy states. The second piece is the Epistemic Head: K duplicated classifier heads trained with an oracle loss that only back-propagates through the best head, with a stop-grad protecting the backbone. Their outputs give a single-forward-pass estimate of epistemic uncertainty, and LTG uses it or entropy or energy as the guiding signal.

What would settle it

Run LTG with increasing guidance weights on a fixed task, and have independent human raters or a separately trained anomaly detector label each generated image as an artifact or a semantically valid longtail example; if the artifact rate rises before correct-class probability drops, or if the FID of $\hat{x}^{(t)}_0$ is worse than the FID of the decoded noisy state at early timesteps, the terminal-estimate premise fails. A complementary check is to match LTG-selected and unguided generations for FID and test whether LTG still produces higher semantic diversity in the production model's feature space.

Watch

Extended reading notes

Core claim

The central claim is that one can generate synthetic training data that is longtail from the perspective of a specific deployed model, not merely diverse from a foundation model's perspective. The method computes a differentiable longtail signal on the decoded terminal-state estimate $\hat{x}^{(t)}_0 = D(P(z_t))$ and biases the denoising update with a term of the form $-w \nabla_{z_t} f^{\mathrm{lt}}_\phi(D(P(z_t))) \sigma_t$. The paper reports that guided generations show lower probability of the correct class, lower accuracy, and higher longtail signal under the production model, while remaining in-distribution according to FID and precision/recall. Training on these generations improves generalization beyond CLIP-space synthetic-data baselines, and the authors attribute the gap to the fact that what is hard for CLIP differs substantially from what is hard for the production model.

Load-bearing premise

The load-bearing premise is that at early denoising steps the one-step terminal-state estimate is already close enough to real images that a classifier trained on clean data gives useful gradients; if the estimate is off-manifold in a way FID misses, guidance may select for artifacts rather than semantic longtail variation.

Editorial extensions

If this is right

  • Across seven natural-image classification benchmarks, fine-tuning with LTG-generated data adds on average 5.6 points of top-1 accuracy over GIF and Dream-ID, and 15.0 points over prompt-tuned Stable Diffusion.
  • On ImageNet-LT, LTG lifts accuracy on the rarest (Few) classes by 10 points over baseline diffusion, while overall accuracy rises 1.3 points, so the benefit is concentrated exactly where data scarcity bites.
  • LTG-generated data are more efficient: the reported gains come with 20 to 30 percent of the synthetic data volume used by GIF, an average efficiency improvement of about 4x.
  • Because neither the diffusion model nor the predictive model is retrained, LTG composes with established training recipes such as logit-adjusted BCE and MAE pretraining, and it can be run iteratively as the model's longtail evolves.
  • Longtail Introspection reduces generated data to text keywords that improve downstream prompting over manual prompt tuning, so the process is inspectable rather than a black-box data factory.

Reading between the lines

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

  • The terminal-state trick is not image-specific: any latent generative model with a differentiable decoder could in principle be steered by a clean-data model, so LTG-style mining could be tested in audio, video, medical imaging, or driving scenes.
  • Because the guiding signal is an arbitrary differentiable function of the production model, one could substitute calibrated loss, safety scores, or downstream task metrics for uncertainty, mining data that are risky rather than merely uncertain.
  • The paper's in-distribution evidence relies on FID and the model's own threshold on correct-class probability; a direct test with an independent density estimator or artifact detector would show whether some guidance weights trade semantic longtail variation for generated artifacts.
  • The vision-language introspection step suggests a closed loop not fully explored: keywords discovered from LTG data could guide real data collection or re-prompt the generator, and the loop could then be repeated, which the paper's saturation results suggest would have diminishing but nonzero returns.
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

4 major / 6 minor

Summary. The paper proposes model-based longtail signals, including a differentiable single-forward-pass Epistemic Head that estimates epistemic uncertainty via multiple heads with an oracle loss, and uses these signals as guidance for a latent diffusion model (Longtail Guidance, LTG). LTG decodes the predicted terminal latent state D(P(z_t)) at each denoising step, evaluates the longtail signal with an existing clean-data classifier, and backpropagates the gradient to bias the denoising estimate toward high-longtail-signal data. The authors claim training-free coupling between off-the-shelf diffusion and a deployed classifier, with large top-1 accuracy gains over GIF, Dream-ID, and prompt-tuned Stable Diffusion across seven benchmarks, and additional VLM-based introspection results. The central results are empirical; no equation reduces to a fitted value, and the main artifacts are external benchmark accuracy numbers.

Significance. If the claims hold, LTG is a practical training-free method to couple existing generative models with deployed classifiers, with potentially large gains on longtail and distribution-shift benchmarks, and the Epistemic Head is a lightweight uncertainty estimator. The paper has several strengths: it evaluates on multiple architectures (ResNet-50, ResNet-34, ViT/LiVT) and multiple datasets; it reports ablations over guidance weight and number of heads (Appendix A.8); it provides a computational-cost analysis (Appendix A.5); and the VLM introspection experiment (Section 4.3, Table 4) is a useful, falsifiable check that LTG data carry semantic value. However, the headline quantitative claim is currently confounded by the generation schedule, and the in-distribution verification is under-specified, so the significance is contingent on additional controlled experiments.

major comments (4)
  1. [Section 4.1, Tables 1 and 2; Supplement A.1, Table 5] The headline comparison to GIF and Dream-ID does not isolate the longtail-guidance mechanism from the iterative generation schedule. LTG generates synthetic data throughout fine-tuning (e.g., Pets at epochs 0,5,...,98; Caltech/Cars/Flowers at epochs 0,5,...,95; ImageNet-100 at 50 images/class every 5 epochs), while GIF and Dream-ID generate all synthetic data once before fine-tuning. Supplement A.1 explicitly states that iteratively generated data outperformed all-at-once generation in the authors' own experiments. Therefore the reported average +5.6 point gain over GIF and Dream-ID could be partly or wholly explained by periodically adding fresh synthetic data during training, rather than by the model-based longtail signal. To support the central claim, the paper needs a control: for example, run LTG with all data generated at epoch 0, and run GIF, Dream-ID, and plain prompt-tuned Stable Diffusion under the same iterative schedule. This concern also applies to the LiVT comparison in Section 4.2 and Table 3 if the LiVT SD baseline generates data only once.
  2. [Section 3.1, Figure 2] The claim that LTG-generated data remain in-distribution is not backed by quantitative evidence. The stated criterion is that P(Class) is lower than unguided generations but 'well above zero'; this uses the same model f_phi that is being guided, so it is not an independent out-of-distribution check. The accompanying statement that FID scores are lower and generative precision+recall are higher than the real-to-real baseline is made without any numbers, table, or figure. Since the central mechanism relies on D(P(z_t)) being close to the real data manifold, please report the actual FID and precision/recall values (with confidence intervals if possible) for the datasets used, and consider an independent OOD detector to verify that guided data do not drift into VAE decoding artifacts.
  3. [Section 3 and Supplement A.7, Figure 15] The justification that an existing clean-data classifier can guide early denoising steps rests on the claim that the decoded terminal state D(P(z_t)) is much closer to real data than the decoded intermediate state D(x_t). This is supported only by a single FID curve for ImageNet-LT, with no numerical values, no confidence intervals, and no analysis of how the FID gap depends on timestep and dataset. Because the gradient in Eq. (11) is computed through D(P(z_t)), a failure mode is that the guidance exploits decoder artifacts rather than semantic longtail variation. Please provide per-dataset quantitative support (e.g., FID/P&R over denoising timesteps, or a distributional test) for the claim that the terminal estimates are sufficiently in-distribution at the guidance timesteps actually used.
  4. [Section 2.1, Eq. (4)] The theoretical justification of the Epistemic Head as approximating posterior predictive samples is asserted rather than derived: head outputs are said to 'act as fixed-point samples from the posterior predictive (Eqn 3)' but no argument or reference is given for why the oracle-loss-trained heads should be interpreted as posterior samples. This is not fatal, because the empirical evaluation in Figure 5 stands on its own, but the wording overclaims a Bayesian justification. Please either provide a derivation or soften the claim to describe the head as a heuristic ensemble approximation, and report error bars or significance tests for the average-precision differences in Figure 5.
minor comments (6)
  1. [Figure 2] The abbreviation P(Class) is used without definition; please define it as the probability of the expected class under the production model f_phi.
  2. [Algorithm 1] The variable z0 is used both for the final latent returned by the algorithm and for the terminal estimate z_hat_0^t; please disambiguate for clarity.
  3. [Section 4.4] There is a typo in 'more than80% generalization improvement'; it should read 'more than 80%'.
  4. [Figure 1 caption] The caption contains a duplicated word: 'the the synthetic data'.
  5. [Supplement A.7] The notation in Figure 13 caption ('D(P (ˆx0 t ))') is inconsistent with the main text's D(P(z_t)); please align the notation.
  6. [Supplement A.5] The text and Table 9 use 'm' for million (e.g., '83.6m parameter VAE'); use 'M' for consistency.

Circularity Check

1 steps flagged · score 2.0 of 10

One Section 3 claim reduces to the LTG objective by construction; the central benchmark results are independent of this step.

  1. self definitional [Section 3, Equation (11) and Figure 2 caption]
    "compute longtail signal f lt ϕ(ˆx 0 t ) ... and then bias the denoising estimate (in latent space) in the direction of higher production model longtail signal ... Guided generations predictably exhibit lower correct class probability, P(Class), lower accuracy, and higher longtail signals,f lt ϕ, compared to unguided generations (zero guidance weight), indicating they are more difficult and more longtail from the predictive model’s perspective."

    Equation (11) defines the denoising update as ˆϵt = ϵθ(zt, t) − w∇zt f lt ϕ (D(P (zt))) σt, i.e. gradient ascent on the model-based longtail signal f lt ϕ evaluated at the decoded terminal estimate. A sample following this update has higher f lt ϕ by construction, so Figure 2's 'higher longtail signal' is the optimized objective restated as an empirical finding rather than an independent prediction. The accompanying 'lower correct class probability' and 'lower accuracy' also follow once f lt includes entropy/total uncertainty, since increasing predictive entropy reduces confidence in every class.

full rationale

The central derivation chain is otherwise self-contained. The Epistemic Head is benchmarked against entropy, energy, and an independently trained ensemble (Figure 5), and the downstream generalization comparisons (Tables 1-4) are measured on public benchmarks with fixed hyperparameters chosen by ablation, so no headline number reduces to a fitted value. I found no load-bearing self-citation: references such as Maddox et al. (2019) (which includes an author) and Hydra/LoRA are background techniques, not uniqueness constraints, and no prior-work theorem is invoked to forbid alternatives. The Supplement A.1 admission that LTG distributes data generation across fine-tuning epochs while GIF/Dream-ID generate once is a scheduling confound, not circularity: it weakens isolation of the mechanism but does not make the measured accuracy gains true by definition. Overall, the only circular ingredient is the Section 3/Figure 2 observation that guided data score higher on the very signal used for guidance; because this tautology is not what establishes the generalization improvements, the paper is only mildly self-definitional.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central method rests on four soft assumptions: one-step terminal estimates are clean enough for a production classifier, Epistemic Head outputs approximate posterior samples, low-but-nonzero predicted probability equals in-distribution, and Stable Diffusion has relevant class knowledge. These are reasonable for an applied paper, but none is formally proven. The main free parameters (guidance weight, number of heads, iterative schedule) are chosen by ablations or by experimental exploration rather than derived. No new physical entities or mediators are introduced.

free parameters (4)
  • Longtail guidance weight w = Energy: 10.0, Entropy: 10.0, Epistemic: 50.0
    Chosen by ablation on Pets (Supplement A.8, Figure 16), then held fixed across classes, datasets, and fine-tuning epochs. The central results in Section 4 depend on this value.
  • Number of Epistemic Heads K = 5
    Chosen by ablation on Pets (Supplement A.8, Figure 17). More heads degrade performance.
  • Use of total uncertainty vs epistemic alone for guidance = epistemic plus aleatoric
    Experimentally found to work better for fine-grained datasets according to Supplement A.1. This signal-choice decision affects all reported results.
  • Synthetic data generation schedule = every 5 epochs over 100 to 300 epochs
    Authors report that iterative generation outperformed generating all at once, but no comparison experiment is shown. The schedule is used in all benchmark comparisons.
assumptions (5)
  • domain assumption One-step DDIM terminal estimate P(z_t) approximates the fully denoised image well enough at early timesteps for a clean-data classifier to score it.
    Section 3, Equation 11, and Supplement A.7: this is the load-bearing premise of LTG; supported empirically by FID, not proven.
  • ad hoc to paper Epistemic Head outputs can be treated as Monte Carlo samples from the posterior predictive distribution p(y|x) in Equation 3.
    Section 2.1 states this after Equation 4; no posterior justification is given beyond initialization prior and oracle diversity loss.
  • domain assumption A low but nonzero predicted class probability is sufficient to keep LTG-generated data in-distribution.
    Section 3.1 uses P(Class) thresholds and reports FID and precision-recall qualitatively, but does not establish that this criterion prevents off-distribution drift.
  • domain assumption Stable Diffusion v1.4 contains enough Internet-scale knowledge to render longtail variations of each target class.
    Used throughout Section 4; if the generative prior is weak for a class, LTG cannot manufacture useful data.
  • domain assumption Top-1 accuracy on the chosen seven datasets is a sufficient proxy for generalization.
    All comparisons and conclusions use top-1 accuracy; no calibration or independent OOD robustness metrics are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative Data Mining with Longtail-Guided Diffusion." pith.science (2026). https://pith.science/paper/KDF5MD67

@misc{pith2026250201980,
  author       = {Pith},
  title        = {Pith review of: Generative Data Mining with Longtail-Guided Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KDF5MD67}},
  note         = {Machine review of arXiv:2502.01980}
}
read the original abstract

It is difficult to anticipate the myriad challenges that a predictive model will encounter once deployed. Common practice entails a reactive, cyclical approach: model deployment, data mining, and retraining. We instead develop a proactive longtail discovery process by imagining additional data during training. In particular, we develop general model-based longtail signals, including a differentiable, single forward pass formulation of epistemic uncertainty that does not impact model parameters or predictive performance but can flag rare or hard inputs. We leverage these signals as guidance to generate additional training data from a latent diffusion model in a process we call Longtail Guidance (LTG). Crucially, we can perform LTG without retraining the diffusion model or the predictive model, and we do not need to expose the predictive model to intermediate diffusion states. Data generated by LTG exhibit semantically meaningful variation, yield significant generalization improvements on numerous image classification benchmarks, and can be analyzed by a VLM to proactively discover, textually explain, and address conceptual gaps in a deployed predictive model.

Figures

Figures reproduced from arXiv: 2502.01980 by the authors.

Figure 1
Figure 1. Diffusion with Longtail Guidance (LTG) generates syn￾thetic data that are difficult or rare for an existing predictive model. The predictive model can be fine-tuned on this data to improve gen￾eralization performance and the the synthetic data can be analyzed to understand conceptual gaps in the predictive model. Guided generations exhibit more extreme views compared to unguided gen￾erations. Results in Section 4 us… view at source ↗
Figure 2
Figure 2. Predictive model performance (left axes, red) and model￾based longtail signals f lt ϕ (right axes, green) on synthetic data generated by varying Longtail Guidance weights for model-based longtail signals: Epistemic (left plot) and Entropy (right plot). Guided generations predictably exhibit lower correct class prob￾ability, P(Class), lower accuracy, and higher longtail signals, f lt ϕ, compared to unguided generatio… view at source ↗
Figure 3
Figure 3. In Longtail Guidance, we iteratively fine-tune existing predictive model fϕ with longtail signal f lt ϕ (left) then freeze model weights ϕ and generate synthetic data with latent diffusion model ϵθ guided by f lt ϕ that is, by definition, rare or disproportionately hard for fϕ (right). Synthetic data are added to fϕ’s training set, and the process repeats. tional compute can be traded for improved performance; see d… view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Epistemic E(y, ϕ) vs aleatoric Eϕ[U(y | ϕ)] uncer￾tainty. Epistemic increases with distance from the data manifold. Aleatoric increases with proximity to the decision boundary. rare in training data. This definition naturally captures a data-centric view in which event…
Figure 5
Figure 5. Figure 5: The Epistemic Head is a better indicator of rare or hard ImageNet-LT validation examples than entropy, energy, or epis￾temic signals from an independently-trained ensemble. f lt ϕ(x) ≈ U 1 K X k p(y | ϕk) ! − 1 K X k U (p(y | ϕk)). (4) Base model performance is protect…
Figure 6
Figure 6. Figure 6: Synthetic data generated for 100 ImageNet classes with increasing Longtail Guidance strength, guided by a SOTA ViT ImageNet￾LT classifier fϕ. Views frequently become more extreme, occluded, or cut off. They also become more difficult for fϕ (see [PITH_FULL_IMAGE:figur…
Figure 7
Figure 7. Figure 7: Universal Guidance (top) vs Longtail Guidance (bottom). Universal Guidance, when driven by model-based longtail signals, successfully raises those signals but it does so by generating data that is no longer in-distribution. It is also much more expensive. Universal Gui…
Figure 8
Figure 8. Figure 8: Longtail keywords from LTG-generated Flowers data. 2024) to caption real training instances xreal and also syn￾thetic data xltg generated by LTG as guided by f lt ϕ . For each synthetic instance, N novel keywords are found by computing the token embeddings that are fur…
Figure 9
Figure 9. Figure 9: Reasoning in latent CLIP space for synthetic data gener￾ation does not account for differences in what is difficult for CLIP versus what is difficult for a production model. We plot (sorted) absolute difference in per-class accuracy between a CLIP zero￾shot classifier …
Figure 10
Figure 10. Figure 10: For fair comparison, we also train on ImageNet-100 and evaluate on the ImageNet-100 (Eval), ImageNet-A, and ImageNet-V2 subsets as defined in (Du et al., 2024). We emphasize that these datasets exhibit example counts that are common for longtail data in production set…
Figure 11
Figure 11. Figure 11: High quantiles of longtail signal indicate examples that are disproportionately hard. 0 1 2 Class 0 1 2 Sample High Epistemic 0 1 2 Class High Aleatoric 0.0 0.2 0.4 0.6 0.8 1.0 [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 13
Figure 13. Figure 13: Performing Longtail Guidance on predicted, decoded terminal states xˆ 0 t (top row) provides production model fϕ with data that are more in-distribution and less corrupted by intermediate diffusion noise than does Naive guidance performed on each intermediate decoded …
Figure 14
Figure 14. Figure 14: Performing Longtail Guidance on the predicted, de￾coded terminal data state xˆ 0 t (top row) provides production model fϕ with cleaner data much earlier in the diffusion denoising pro￾cess, enabling it to meaningfully exercise guidance as compared to naively performin…
Figure 17
Figure 17. Figure 17: Ablation over Epistemic Heads on Predictive Model Generalization Performance on the Pets Dataset Expan￾sion Task, holding the guidance weight fixed at 50.0 [PITH_FULL_IMAGE:figures/full_fig_p020_17.png]
Figure 18
Figure 18. Figure 18: Additional example of baseline diffusion (left) and Longtail Guidance (right). Baseline diffusion tends to generate canonical, well-posed views that help a predictive model generalize, but only up to a point. In contrast, Longtail Guidance produces more extreme, occlu…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

104 extracted references · 39 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Introducing the next generation of claude, 2024

    Anthropic. Introducing the next generation of claude, 2024. URL https://www.anthropic.com/news/claude-3-family. Accessed: 2025-01-03

    2024

  3. [3]

    Azizi, S., Kornblith, S., Saharia, C., Norouzi, M., and Fleet, D. J. Synthetic data from diffusion models improves imagenet classification. arXiv preprint arXiv:2304.08466, 2023

    arXiv 2023

  4. [4]

    Universal guidance for diffusion models

    Bansal, A., Chu, H.-M., Schwarzschild, A., Sengupta, S., Goldblum, M., Geiping, J., and Goldstein, T. Universal guidance for diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 843--852, 2023

    2023

  5. [5]

    and Grover, A

    Bansal, H. and Grover, A. Leaving reality to imagination: Robust classification via generated datasets. arXiv preprint arXiv:2302.02503, 2023

    arXiv 2023

  6. [6]

    Beit: Bert pre-training of image transformers

    Bao, H., Dong, L., Piao, S., and Wei, F. Beit: Bert pre-training of image transformers. arXiv preprint arXiv:2106.08254, 2021

    arXiv 2021

  7. [7]

    D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. Language models are few-shot learners. Advances in neural information processing systems, 33: 0 1877--1901, 2020

    1901

  8. [8]

    Ace: Ally complementary experts for solving long-tailed recognition in one-shot

    Cai, J., Wang, Y., and Hwang, J.-N. Ace: Ally complementary experts for solving long-tailed recognition in one-shot. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 112--121, 2021

    2021

Show all 104 references
  1. [9]

    Learning imbalanced datasets with label-distribution-aware margin loss

    Cao, K., Wei, C., Gaidon, A., Arechiga, N., and Ma, T. Learning imbalanced datasets with label-distribution-aware margin loss. Advances in neural information processing systems, 32, 2019

    2019

  2. [10]

    Pixart- : Weak-to-strong training of diffusion transformer for 4k text-to-image generation

    Chen, J., Ge, C., Xie, E., Wu, Y., Yao, L., Ren, X., Wang, Z., Luo, P., Lu, H., and Li, Z. Pixart- : Weak-to-strong training of diffusion transformer for 4k text-to-image generation. In European Conference on Computer Vision, pp.\ 74--91. Springer, 2025

    2025

  3. [11]

    Gridmask data augmentation

    Chen, P., Liu, S., Zhao, H., Wang, X., and Jia, J. Gridmask data augmentation. arXiv preprint arXiv:2001.04086, 2020

    2001 arXiv

  4. [12]

    Detecting adversarial samples using influence functions and nearest neighbors

    Cohen, G., Sapiro, G., and Giryes, R. Detecting adversarial samples using influence functions and nearest neighbors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 14453--14462, 2020

    2020

  5. [13]

    A., Birch, A., Abraham, T

    Crowson, K., Baumann, S. A., Birch, A., Abraham, T. M., Kaplan, D. Z., and Shippole, E. Scalable high-resolution pixel-space image synthesis with hourglass diffusion transformers. In Forty-first International Conference on Machine Learning, 2024

    2024

  6. [14]

    D., Zoph, B., Mane, D., Vasudevan, V., and Le, Q

    Cubuk, E. D., Zoph, B., Mane, D., Vasudevan, V., and Le, Q. V. Autoaugment: Learning augmentation strategies from data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 113--123, 2019

    2019

  7. [15]

    D., Zoph, B., Shlens, J., and Le, Q

    Cubuk, E. D., Zoph, B., Shlens, J., and Le, Q. V. Randaugment: Practical automated data augmentation with a reduced search space. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pp.\ 702--703, 2020

    2020

  8. [16]

    Parametric contrastive learning

    Cui, J., Zhong, Z., Liu, S., Yu, B., and Jia, J. Parametric contrastive learning. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 715--724, 2021

    2021

  9. [17]

    Class-balanced loss based on effective number of samples

    Cui, Y., Jia, M., Lin, T.-Y., Song, Y., and Belongie, S. Class-balanced loss based on effective number of samples. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 9268--9277, 2019

    2019

  10. [18]

    Decomposition of uncertainty in bayesian deep learning for efficient and risk-sensitive learning

    Depeweg, S., Hernandez-Lobato, J.-M., Doshi-Velez, F., and Udluft, S. Decomposition of uncertainty in bayesian deep learning for efficient and risk-sensitive learning. In International conference on machine learning, pp.\ 1184--1193. PMLR, 2018

    2018

  11. [19]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Devlin, J. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  12. [20]

    Improved regularization of convolutional neural networks with cutout

    DeVries, T. Improved regularization of convolutional neural networks with cutout. arXiv preprint arXiv:1708.04552, 2017

    2017 arXiv

  13. [21]

    and Nichol, A

    Dhariwal, P. and Nichol, A. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34: 0 8780--8794, 2021

    2021

  14. [22]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Dosovitskiy, A. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  15. [23]

    and Song, Y

    Dou, Z. and Song, Y. Diffusion posterior sampling for linear inverse problem solving: A filtering perspective. In The Twelfth International Conference on Learning Representations, 2024

    2024

  16. [24]

    Dream the impossible: Outlier imagination with diffusion models

    Du, X., Sun, Y., Zhu, J., and Li, Y. Dream the impossible: Outlier imagination with diffusion models. Advances in Neural Information Processing Systems, 36, 2024

    2024

  17. [25]

    Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories

    Fei-Fei, L., Fergus, R., and Perona, P. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. In 2004 conference on computer vision and pattern recognition workshop, pp.\ 178--178. IEEE, 2004

    2004

  18. [26]

    and Ghahramani, Z

    Gal, Y. and Ghahramani, Z. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In international conference on machine learning, pp.\ 1050--1059. PMLR, 2016

    2016

  19. [27]

    J., and Tao, D

    Gou, J., Yu, B., Maybank, S. J., and Tao, D. Knowledge distillation: A survey. International Journal of Computer Vision, 129 0 (6): 0 1789--1819, 2021

    2021

  20. [28]

    A., and Mann, T

    Gowal, S., Rebuffi, S.-A., Wiles, O., Stimberg, F., Calian, D. A., and Mann, T. A. Improving robustness using generated data. Advances in Neural Information Processing Systems, 34: 0 4218--4233, 2021

    2021

  21. [29]

    Multiple choice learning: Learning to produce multiple structured outputs

    Guzman-Rivera, A., Batra, D., and Kohli, P. Multiple choice learning: Learning to produce multiple structured outputs. Advances in neural information processing systems, 25, 2012

    2012

  22. [30]

    Masked autoencoders are scalable vision learners

    He, K., Chen, X., Xie, S., Li, Y., Doll \'a r, P., and Girshick, R. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 16000--16009, 2022

    2022

  23. [31]

    D., Zoph, B., Gilmer, J., and Lakshminarayanan, B

    Hendrycks, D., Mu, N., Cubuk, E. D., Zoph, B., Gilmer, J., and Lakshminarayanan, B. Augmix: A simple data processing method to improve robustness and uncertainty. arXiv preprint arXiv:1912.02781, 2019

    1912 arXiv

  24. [32]

    The many faces of robustness: A critical analysis of out-of-distribution generalization

    Hendrycks, D., Basart, S., Mu, N., Kadavath, S., Wang, F., Dorundo, E., Desai, R., Zhu, T., Parajuli, S., Guo, M., et al. The many faces of robustness: A critical analysis of out-of-distribution generalization. In Proceedings of the IEEE/CVF international conference on compute...

    2021

  25. [33]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017

    2017

  26. [34]

    and Salimans, T

    Ho, J. and Salimans, T. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022

    2022 arXiv

  27. [35]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  28. [36]

    Safa: Sample-adaptive feature augmentation for long-tailed image classification

    Hong, Y., Zhang, J., Sun, Z., and Yan, K. Safa: Sample-adaptive feature augmentation for long-tailed image classification. In European Conference on Computer Vision, pp.\ 587--603. Springer, 2022

    2022

  29. [37]

    Generalized odin: Detecting out-of-distribution image without learning from out-of-distribution data

    Hsu, Y.-C., Shen, Y., Jin, H., and Kira, Z. Generalized odin: Detecting out-of-distribution image without learning from out-of-distribution data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10951--10960, 2020

    2020

  30. [38]

    J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W

    Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021

    2021 arXiv

  31. [39]

    On the importance of gradients for detecting distributional shifts in the wild

    Huang, R., Geng, A., and Li, Y. On the importance of gradients for detecting distributional shifts in the wild. Advances in Neural Information Processing Systems, 34: 0 677--689, 2021

    2021

  32. [40]

    Categorical reparameterization with gumbel-softmax

    Jang, E., Gu, S., and Poole, B. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144, 2016

    2016 arXiv

  33. [41]

    Motiondiffuser: Controllable multi-agent motion prediction using diffusion

    Jiang, C., Cornman, A., Park, C., Sapp, B., Zhou, Y., Anguelov, D., et al. Motiondiffuser: Controllable multi-agent motion prediction using diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 9644--9653, 2023

    2023

  34. [42]

    M., Najibi, M., Qi, C

    Jiang, C. M., Najibi, M., Qi, C. R., Zhou, Y., and Anguelov, D. Improving the intra-class long-tail in 3d detection via rare example mining. In European Conference on Computer Vision, pp.\ 158--175. Springer, 2022

    2022

  35. [43]

    Decoupling representation and classifier for long-tailed recognition

    Kang, B., Xie, S., Rohrbach, M., Yan, Z., Gordo, A., Feng, J., and Kalantidis, Y. Decoupling representation and classifier for long-tailed recognition. arXiv preprint arXiv:1910.09217, 2019

    1910 arXiv

  36. [44]

    B., Chess, B., Child, R., Gray, S., Radford, A., Wu, J., and Amodei, D

    Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B., Chess, B., Child, R., Gray, S., Radford, A., Wu, J., and Amodei, D. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361, 2020

    2001 arXiv

  37. [45]

    Elucidating the design space of diffusion-based generative models

    Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of diffusion-based generative models. Advances in neural information processing systems, 35: 0 26565--26577, 2022

    2022

  38. [46]

    Collecting a large-scale dataset of fine-grained cars

    Krause, J., Deng, J., Stark, M., and Fei-Fei, L. Collecting a large-scale dataset of fine-grained cars. 2013

    2013

  39. [47]

    Multi-concept customization of text-to-image diffusion

    Kumari, N., Zhang, B., Zhang, R., Shechtman, E., and Zhu, J.-Y. Multi-concept customization of text-to-image diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 1931--1941, 2023

    1931

  40. [48]

    Improved precision and recall metric for assessing generative models

    Kynk \"a \"a nniemi, T., Karras, T., Laine, S., Lehtinen, J., and Aila, T. Improved precision and recall metric for assessing generative models. Advances in neural information processing systems, 32, 2019

    2019

  41. [49]

    Simple and scalable predictive uncertainty estimation using deep ensembles

    Lakshminarayanan, B., Pritzel, A., and Blundell, C. Simple and scalable predictive uncertainty estimation using deep ensembles. Advances in neural information processing systems, 30, 2017

    2017

  42. [50]

    A simple unified framework for detecting out-of-distribution samples and adversarial attacks

    Lee, K., Lee, K., Lee, H., and Shin, J. A simple unified framework for detecting out-of-distribution samples and adversarial attacks. Advances in neural information processing systems, 31, 2018

    2018

  43. [51]

    Trustworthy long-tailed classification

    Li, B., Han, Z., Li, H., Fu, H., and Zhang, C. Trustworthy long-tailed classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6970--6979, 2022 a

    2022

  44. [52]

    W., Kreis, K., Fidler, S., and Torralba, A

    Li, D., Ling, H., Kim, S. W., Kreis, K., Fidler, S., and Torralba, A. Bigdatasetgan: Synthesizing imagenet with pixel-wise annotations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 21330--21340, 2022 b

    2022

  45. [53]

    Nested collaborative learning for long-tailed visual recognition

    Li, J., Tan, Z., Wan, J., Lei, Z., and Guo, G. Nested collaborative learning for long-tailed visual recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6949--6958, 2022 c

    2022

  46. [54]

    Long-tailed visual recognition via gaussian clouded logit adjustment

    Li, M., Cheung, Y.-m., and Lu, Y. Long-tailed visual recognition via gaussian clouded logit adjustment. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6929--6938, 2022 d

    2022

  47. [55]

    S., Indyk, P., and Katabi, D

    Li, T., Cao, P., Yuan, Y., Fan, L., Yang, Y., Feris, R. S., Indyk, P., and Katabi, D. Targeted supervised contrastive learning for long-tailed recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 6918--6928, 2022 e

    2022

  48. [56]

    Liu, H., Li, C., Li, Y., Li, B., Zhang, Y., Shen, S., and Lee, Y. J. Llava-next: Improved reasoning, ocr, and world knowledge, January 2024. URL https://llava-vl.github.io/blog/2024-01-30-llava-next/

    2024

  49. [57]

    Accurate uncertainty estimation and decomposition in ensemble learning

    Liu, J., Paisley, J., Kioumourtzoglou, M.-A., and Coull, B. Accurate uncertainty estimation and decomposition in ensemble learning. Advances in neural information processing systems, 32, 2019 a

    2019

  50. [58]

    Energy-based out-of-distribution detection

    Liu, W., Wang, X., Owens, J., and Li, Y. Energy-based out-of-distribution detection. Advances in neural information processing systems, 33: 0 21464--21475, 2020

    2020

  51. [59]

    Liu, Z., Miao, Z., Zhan, X., Wang, J., Gong, B., and Yu, S. X. Large-scale long-tailed recognition in an open world. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 2537--2546, 2019 b

    2019

  52. [60]

    Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models

    Lu, C., Zhou, Y., Bao, F., Chen, J., Li, C., and Zhu, J. Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models. arXiv preprint arXiv:2211.01095, 2022

    2022 arXiv

  53. [61]

    J., Izmailov, P., Garipov, T., Vetrov, D

    Maddox, W. J., Izmailov, P., Garipov, T., Vetrov, D. P., and Wilson, A. G. A simple baseline for bayesian uncertainty in deep learning. Advances in neural information processing systems, 32, 2019

    2019

  54. [62]

    The Emotion Machine: Commonsense Thinking, Artificial Intelligence, and the Future of the Human Mind

    Minsky, M. The Emotion Machine: Commonsense Thinking, Artificial Intelligence, and the Future of the Human Mind. Simon and Schuster, 2006. ISBN 9780743276641

    2006

  55. [63]

    and Zisserman, A

    Nilsback, M.-E. and Zisserman, A. Automated flower classification over a large number of classes. In 2008 Sixth Indian conference on computer vision, graphics & image processing, pp.\ 722--729. IEEE, 2008

    2008

  56. [64]

    Gpt-4 technical report

    OpenAI. Gpt-4 technical report. Technical report, OpenAI, 2023. URL https://openai.com/research/gpt-4

    2023

  57. [65]

    M., Vedaldi, A., Zisserman, A., and Jawahar, C

    Parkhi, O. M., Vedaldi, A., Zisserman, A., and Jawahar, C. Cats and dogs. In 2012 IEEE conference on computer vision and pattern recognition, pp.\ 3498--3505. IEEE, 2012

    2012

  58. [66]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., M \"u ller, J., Penna, J., and Rombach, R. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023

    2023 arXiv

  59. [67]

    R., Hendy, N., Wang, Z., Morales, A., Wang, K., and Roy, N

    Pronovost, E., Ganesina, M. R., Hendy, N., Wang, Z., Morales, A., Wang, K., and Roy, N. Scenario diffusion: Controllable driving scenario generation with diffusion. Advances in Neural Information Processing Systems, 36: 0 68873--68894, 2023

    2023

  60. [68]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pp.\ 8748--8763. PMLR, 2021

    2021

  61. [69]

    Towards realistic scene generation with lidar diffusion models

    Ran, H., Guizilini, V., and Wang, Y. Towards realistic scene generation with lidar diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 14738--14748, 2024

    2024

  62. [70]

    High-resolution image synthesis with latent diffusion models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10684--10695, 2022

    2022

  63. [71]

    and Doll \'a r, G

    Ross, T.-Y. and Doll \'a r, G. Focal loss for dense object detection. In proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 2980--2988, 2017

    2017

  64. [72]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Ruiz, N., Li, Y., Jampani, V., Pritch, Y., Rubinstein, M., and Aberman, K. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 22500--22510, 2023

    2023

  65. [73]

    Denoising diffusion implicit models

    Song, J., Meng, C., and Ermon, S. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020 a

    2010 arXiv

  66. [74]

    Loss-guided diffusion models for plug-and-play controllable generation

    Song, J., Zhang, Q., Yin, H., Mardani, M., Liu, M.-Y., Kautz, J., Chen, Y., and Vahdat, A. Loss-guided diffusion models for plug-and-play controllable generation. In International Conference on Machine Learning, pp.\ 32483--32498. PMLR, 2023

    2023

  67. [75]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020 b

    2011 arXiv

  68. [76]

    Long-tailed classification by keeping the good and removing the bad momentum causal effect

    Tang, K., Huang, J., and Zhang, H. Long-tailed classification by keeping the good and removing the bad momentum causal effect. Advances in neural information processing systems, 33: 0 1513--1524, 2020

    2020

  69. [77]

    Deit iii: Revenge of the vit

    Touvron, H., Cord, M., and J \'e gou, H. Deit iii: Revenge of the vit. In European conference on computer vision, pp.\ 516--533. Springer, 2022

    2022

  70. [78]

    S., Roth, K., Swiatkowski, J., Dillon, J

    Tran, L., Veeling, B. S., Roth, K., Swiatkowski, J., Dillon, J. V., Snoek, J., Mandt, S., Salimans, T., Nowozin, S., and Jenatton, R. Hydra: Preserving ensemble diversity for model distillation. arXiv preprint arXiv:2001.04694, 2020

    2001 arXiv

  71. [79]

    and Ye, J

    Um, S. and Ye, J. C. Self-guided generation of minority samples using diffusion models. In European Conference on Computer Vision, pp.\ 414--430. Springer, 2024

    2024

  72. [80]

    W., and Gal, Y

    Van Amersfoort, J., Smith, L., Teh, Y. W., and Gal, Y. Uncertainty estimation using a single deep deterministic neural network. In International conference on machine learning, pp.\ 9690--9700. PMLR, 2020

    2020

  73. [81]

    The inaturalist species classification and detection dataset

    Van Horn, G., Mac Aodha, O., Song, Y., Cui, Y., Sun, C., Shepard, A., Adam, H., Perona, P., and Belongie, S. The inaturalist species classification and detection dataset. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 8769--8778, 2018

    2018

  74. [82]

    End-to-end diffusion latent optimization improves classifier guidance

    Wallace, B., Gokul, A., Ermon, S., and Naik, N. End-to-end diffusion latent optimization improves classifier guidance. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 7280--7290, 2023 a

    2023

  75. [83]

    Edict: Exact diffusion inversion via coupled transformations

    Wallace, B., Gokul, A., and Naik, N. Edict: Exact diffusion inversion via coupled transformations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 22532--22541, 2023 b

    2023

  76. [84]

    Towards calibrated hyper-sphere representation via distribution overlap coefficient for long-tailed learning

    Wang, H., Fu, S., He, X., Fang, H., Liu, Z., and Hu, H. Towards calibrated hyper-sphere representation via distribution overlap coefficient for long-tailed learning. In European Conference on Computer Vision, pp.\ 179--196. Springer, 2022

    2022

  77. [85]

    Wang, X., Lian, L., Miao, Z., Liu, Z., and Yu, S. X. Long-tailed recognition by routing diverse distribution-aware experts. arXiv preprint arXiv:2010.01809, 2020

    2010 arXiv

  78. [86]

    V., Zhou, D., et al

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q. V., Zhou, D., et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 0 24824--24837, 2022

    2022

  79. [87]

    Wilson, A. G. and Izmailov, P. Bayesian deep learning and a probabilistic perspective of generalization. Advances in neural information processing systems, 33: 0 4697--4708, 2020

    2020

  80. [88]

    Learning from multiple experts: Self-paced knowledge distillation for long-tailed classification

    Xiang, L., Ding, G., and Han, J. Learning from multiple experts: Self-paced knowledge distillation for long-tailed classification. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part V 16, pp.\ 247--263. Springer, 2020

    2020

  81. [89]

    Constructing balance from imbalance for long-tailed image recognition

    Xu, Y., Li, Y.-L., Li, J., and Lu, C. Constructing balance from imbalance for long-tailed image recognition. In European Conference on Computer Vision, pp.\ 38--56. Springer, 2022

    2022

  82. [90]

    Learning imbalanced data with vision transformers

    Xu, Z., Liu, R., Yang, S., Chai, Z., and Yuan, C. Learning imbalanced data with vision transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 15793--15803, 2023

    2023

  83. [91]

    Tree of thoughts: Deliberate problem solving with large language models

    Yao, S., Yu, D., Zhao, J., Shafran, I., Griffiths, T., Cao, Y., and Narasimhan, K. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36, 2024

    2024

  84. [92]

    Bdd100k: A diverse driving dataset for heterogeneous multitask learning

    Yu, F., Chen, H., Wang, X., Xian, W., Chen, Y., Liu, F., Madhavan, V., and Darrell, T. Bdd100k: A diverse driving dataset for heterogeneous multitask learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 2636--2645, 2020

    2020

  85. [93]

    Freedom: Training-free energy-guided conditional diffusion model

    Yu, J., Wang, Y., Zhao, C., Ghanem, B., and Zhang, J. Freedom: Training-free energy-guided conditional diffusion model. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 23174--23184, 2023 a

    2023

  86. [94]

    Dataset distillation: A comprehensive review

    Yu, R., Liu, S., and Wang, X. Dataset distillation: A comprehensive review. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023 b

    2023

  87. [95]

    J., Chun, S., Choe, J., and Yoo, Y

    Yun, S., Han, D., Oh, S. J., Chun, S., Choe, J., and Yoo, Y. Cutmix: Regularization strategy to train strong classifiers with localizable features. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 6023--6032, 2019

    2019

  88. [96]

    Adding conditional control to text-to-image diffusion models

    Zhang, L., Rao, A., and Agrawala, M. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 3836--3847, 2023 a

    2023

  89. [97]

    Memo: Test time robustness via adaptation and augmentation

    Zhang, M., Levine, S., and Finn, C. Memo: Test time robustness via adaptation and augmentation. Advances in neural information processing systems, 35: 0 38629--38642, 2022 a

    2022

  90. [98]

    Distribution alignment: A unified framework for long-tail visual recognition

    Zhang, S., Li, Z., Yan, S., He, X., and Sun, J. Distribution alignment: A unified framework for long-tail visual recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 2361--2370, 2021

    2021

  91. [99]

    Self-supervised aggregation of diverse experts for test-agnostic long-tailed recognition

    Zhang, Y., Hooi, B., Hong, L., and Feng, J. Self-supervised aggregation of diverse experts for test-agnostic long-tailed recognition. Advances in Neural Information Processing Systems, 35: 0 34077--34090, 2022 b

    2022

  92. [100]

    Expanding small-scale datasets with guided imagination

    Zhang, Y., Zhou, D., Hooi, B., Wang, K., and Feng, J. Expanding small-scale datasets with guided imagination. Advances in neural information processing systems, 36: 0 76558--76618, 2023 b

    2023

  93. [101]

    Improving calibration for long-tailed recognition

    Zhong, Z., Cui, J., Liu, S., and Jia, J. Improving calibration for long-tailed recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 16489--16498, 2021

    2021

  94. [102]

    Language-guided traffic simulation via scene-level diffusion

    Zhong, Z., Rempe, D., Chen, Y., Ivanovic, B., Cao, Y., Xu, D., Pavone, M., and Ray, B. Language-guided traffic simulation via scene-level diffusion. In Conference on Robot Learning, pp.\ 144--177. PMLR, 2023

    2023

  95. [103]

    Training on thin air: Improve image classification with generated data

    Zhou, Y., Sahak, H., and Ba, J. Training on thin air: Improve image classification with generated data. arXiv preprint arXiv:2305.15316, 2023

    2023 arXiv

  96. [104]

    P., and Jiang, Y.-G

    Zhu, J., Wang, Z., Chen, J., Chen, Y.-P. P., and Jiang, Y.-G. Balanced contrastive learning for long-tailed visual recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6908--6917, 2022

    2022

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.