Uncovering Memorization Effect in the Presence of Spurious Correlations

Chenyu You; Haocheng Dai; James S. Duncan; Jasjeet S. Sekhon; Sarang Joshi; Yifei Min

REVIEW 3 major objections 4 minor 24 references

Uncovering Memorization Effect in the Presence of Spurious Correlations

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that spurious memorization, concentrated in a tiny subset of neurons, drives worst-group accuracy gaps, and that pruning 0.01% of those neurons during fine-tuning substantially closes those gaps.

desk verdict The descriptive finding on critical neurons is solid and worth taking seriously; the fine-tuning story needs controls before the mechanistic claim lands. read the letter →

arxiv 2501.00961 v3 pith:CXRWUJE4 submitted 2025-01-01 cs.LG cs.AIcs.CVeess.IV

Chenyu You , Haocheng Dai , Yifei Min , Jasjeet S. Sekhon , Sarang Joshi , James S. Duncan This is my paper · ORCID

classification cs.LGcs.AIcs.CVeess.IV

keywords spuriouscorrelationsmemorizationcriticalneuronsworst-groupaccuracygrouprobustnessneuralnetworkpruningcontrastivelearning

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 argues that when models trained on spurious correlations perform badly on minority groups, the root cause is memorization: a small set of critical neurons learns to predict atypical (minority) training examples, so minority accuracy is high in training but collapses at test time because it depends on a few fragile neurons. It establishes this by showing that zeroing out one to three of these neurons drops minority training accuracy by up to 11.96% while majority groups drop at most 2.15%. It then proposes a fine-tuning framework that prunes 0.01% of neurons and contrasts the pruned model with the target model, raising worst-group test accuracy from 64.0% to 80.9% on Waterbirds/ResNet-50 and from 52.7% to 76.5% on Waterbirds/ViT-Small. If correct, this means spurious-correlation failure is localized in a tiny fraction of parameters and can be mitigated with a very small intervention.

What carries the argument

The central object is the 'critical neuron': a channel (in ResNet-50) or neuron (in ViT-Small) whose removal disproportionately affects minority group accuracy. They are identified by two criteria: gradient norm (group-specific loss gradient) and weight magnitude (group-invariant). The mechanism that carries the argument is a dual-branch fine-tuning framework: an auxiliary model is created by masking the top 0.01% of neurons by gradient or magnitude, and the target model is trained to align its features with the auxiliary model using NT-Xent contrastive loss plus MSE classification loss, so the model learns not to rely on the spurious memorization encapsulated in those neurons.

What would settle it

Run the same fine-tuning framework on Waterbirds/ResNet-50 with no pruning, or with 0.01% of neurons chosen at random for the auxiliary model; if worst-group accuracy still rises by roughly 17 points over ERM, then the identified critical neurons are not the cause of the improvement.

Watch

Extended reading notes

Core claim

On the paper's own terms: neural networks trained by empirical risk minimization with spurious correlations store minority-group information in a very small set of 'critical neurons'. Evidence comes from three sources: unstructured tracing (zeroing top-k neurons by gradient or magnitude across the whole network), structured tracing (same within each layer), and perturbation experiments (random initialization and random noise). All three show minority-group training accuracy is disproportionately sensitive to these neurons, while majority groups are barely affected. The paper further claims that contrasting the target model with a pruned auxiliary model during fine-tuning removes this spurious memorization and improves worst-group accuracy.

Load-bearing premise

The load-bearing assumption is that the accuracy gains come specifically from removing the identified critical neurons, rather than from the fine-tuning procedure or from pruning any random neurons; the paper does not run a control that applies the same contrastive fine-tuning without pruning or with randomly selected neurons.

Editorial extensions

If this is right

Worst-group robustness can be improved by intervening on 0.01% of parameters, suggesting that spurious memorization is highly localized rather than diffuse.
The critical-neuron criterion (gradient norm or weight magnitude) provides a cheap diagnostic for which groups are memorized rather than generalized.
Contrasting a model with its pruned twin can serve as a training signal that discourages reliance on a few neurons, without needing group labels for every sample.
The reported gains on Waterbirds and CelebA across ResNet-50 and ViT-Small indicate the mechanism may transfer across architectures and spurious-correlation benchmarks.

Reading between the lines

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

If the mechanism generalizes, measuring the concentration of critical neurons (e.g., how few neurons carry minority accuracy) could predict a model's vulnerability to spurious correlations before test-time evaluation.
The paper's gradient-based criterion evaluates neurons on the same group whose accuracy they affect; a stricter test would check whether neurons identified on one minority group also matter for another, which would separate memorization from generic feature importance.
No control runs the fine-tuning framework with zero pruning or random pruning, so part of the reported WGA gains may be attributable to contrastive fine-tuning itself rather than specifically to removing critical neurons.
The 0.01% pruning ratio is surprisingly small; testing whether the optimal ratio scales with dataset size or number of minority examples would clarify whether the memorization set is truly fixed or grows with data.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: the descriptive result is genuinely interesting and reasonably supported. A small set of neurons — channels in ResNet, units in ViT — can be identified by gradient or magnitude such that zeroing them drops minority-group training accuracy by up to ~10-12% while majority groups move a couple of points. The magnitude criterion is group-agnostic, and the random-initialization/noise ablations plus the train/test contrast in Tables 13–14 strengthen the memorization reading: modifying these neurons moves training accuracy much more than test accuracy. That is a real empirical phenomenon, and it is not something I had seen demonstrated cleanly in the spurious-correlation setting before.

The new twist is the two-stage framing: trace critical neurons, then use them to define an auxiliary pruned branch in a contrastive fine-tuning loss. The WGA gains are large (80.9 vs 64.0 on Waterbirds/ResNet-50; 76.5 vs 52.7 on ViT). But this is where the paper goes soft. The fine-tuning changes several things at once: continued training, an MSE term, an NT-Xent alignment to an auxiliary model, hard-example selection, and an adaptively recomputed 0.01% mask. No control with a random mask of the same density, and no control with an unpruned auxiliary. Without those, you cannot attribute the WGA improvement to removing the identified critical neurons. It could be a generic regularization effect of the dual-branch contrastive objective. The ablations in Tables 4–10 tune hyperparameters but do not isolate the mask as the causal ingredient. That is a load-bearing gap for the mechanistic claim, not a cosmetic one.

Minor issues: the main figures lack error bars (Table 12 gives means±std, which helps), and the gradient-based tracing criterion is partly circular because it selects neurons by a group's gradient and then measures that same group's accuracy. The magnitude criterion mitigates this, so I would not call it fatal. Also, the 'first-ever evidence' language in the abstract is over-claimed; the paper extends existing localization results (Maini et al. 2023, Stephenson et al. 2021) rather than inventing the concept.

Who this is for: people working on spurious correlations, memorization, and interpretability. The descriptive tracing result is worth engaging with seriously. The fine-tuning method needs a revision with the missing controls before I would believe the causal story. But that is fixable, and the paper deserves a serious referee. I would send it out rather than desk-reject.

Referee Report

3 major / 4 minor

Summary. This paper studies the role of memorization in spurious-correlation group imbalance. Using Waterbirds and CelebA with ResNet-50 and ViT-Small, the authors define 'spurious memorization' as the concentration of minority-group training accuracy in a small set of critical neurons. In Stage I (Section 2.1), gradient- and magnitude-based tracing identify and prune or perturb top-k neurons, showing that minority group training accuracy drops substantially (up to 11.96%) while majority groups drop at most 2.15%; random-initialization and noise ablations plus Tables 13-14 support a memorization interpretation. In Stage II (Section 2.2), a fine-tuning framework contrasts a target model with an adaptively pruned auxiliary model under NT-Xent and MSE losses, reporting worst-group accuracy gains from 64.0 to 80.9 on Waterbirds/ResNet-50 and from 52.7 to 76.5 on Waterbirds/ViT-Small. The paper concludes that removing spurious memorization of critical neurons causes the robustness improvement.

Significance. The descriptive finding is valuable and, in part, well supported: the magnitude-based criterion is group-invariant, the random-initialization and noise controls show that minority training accuracy is unusually sensitive to a few neurons, and the train/test discrepancy in Tables 13-14 supports a memorization rather than a generalization role for those neurons. If the causal claim were established, the paper would be a strong contribution to mechanistic accounts of group robustness. However, the causal claim is not yet established, because the Stage II framework varies several components jointly and lacks controls for random masks or an unpruned auxiliary. The code link is included, which is useful for reproducibility, but the missing intervention controls are the main barrier to the paper's central conclusion.

major comments (3)

[Section 2.2.1-2.2.2, Eq. (2.3)] The worst-group accuracy improvements in Figure 6 cannot be attributed to removal of the identified critical neurons, because the comparison is only against ERM. The proposed fine-tuning simultaneously changes the training schedule, replaces cross-entropy with MSE, adds an NT-Xent contrastive term against an auxiliary branch, selects hard examples for gradient computation, and adaptively recomputes a 0.01% mask each epoch. There is no control with an unpruned auxiliary branch, no control with a same-density random mask redrawn each epoch, and no control with a fixed non-adaptive mask. If a random-mask condition reproduces the 80.9% or 76.5% WGA, the result would be explained by the dual-branch contrastive regularizer rather than by critical-neuron memorization. Since the central causal claim in Section 3 rests on this experiment, these controls are required before the claim can be accepted.
[Section 2.1.1, Figure 2] The gradient-based tracing criterion is partially circular: neurons are selected as the top-gradient neurons for group j and then evaluated on the same group j. The magnitude criterion does not share this issue, and Figure 3 shows overlap between the two sets, so the descriptive claim survives; nevertheless, the textual claim that 'both gradient and magnitude-based criteria are effective' should be qualified, and the authors should present the magnitude-based results as the primary non-circular evidence for critical-neuron existence.
[Section 2.2.1, Table 6] The ablation in Table 6 shows that 10 fine-tuning epochs yields WGA 54.98%, which is below ERM, while 20 and 30 epochs yield 80.89%. This large and non-monotonic dependence on fine-tuning length suggests that the reported gains are not solely a simple function of the mask; without disclosing the checkpoint-selection details, this sensitivity undermines the attribution of the final gain to spurious memorization removal and should be discussed explicitly.

minor comments (4)

[Section 4] The Methods text states that 'all accuracy metrics reported in this paper are derived from the test set', which contradicts the explicit statement in Section 2.1 that all group accuracy evaluated before and after pruning is evaluated on the training set; please clarify which numbers are train and which are test.
[Eq. (2.1)] In the definition of cosine similarity, the denominator reads 'u· v/(||u||·|| b||)' where b is undefined; it should presumably be v.
[Figure 9 caption] The caption contains a typo: 'L NE-Xent' should be 'NT-Xent'.
[Section 2.2.1, Table 9] The paper says 0.01% pruning percentage serves as a hyperparameter, but Table 9 shows extreme sensitivity (10% gradient-based pruning drops WGA to 0.04%); a brief discussion of this sensitivity and its implications for the 0.01% default would strengthen the presentation.

Circularity Check

1 steps flagged · score 4.0 of 10

Gradient-based tracing selects neurons by a group's own gradient and then measures that same group's accuracy, a partial validation loop; the magnitude criterion and Stage I controls provide independent support, while Stage II lacks a random-mask control but is not circular by construction.

self definitional [Section 2.1.1 (Unstructured Tracing), equations defining v(i,j) and Δacc(j)]
"For each group j, we select those neurons i′s of which the ∥v(i,j)∥2 are the top-k largest among all M neurons... The change in accuracy ∆acc for each group j is given by ∆acc(j) =|acc(Dj,f (θ,·))− acc(Dj,f (mj⊙ θ,·))|."

v(i,j) is the gradient of the cross-entropy loss on group Dj with respect to neuron zi, and the same group Dj is used both to select the top-k neurons and to measure the accuracy change. Removing neurons with the largest per-group loss gradient is a direct sensitivity operation on that group's training loss, so a subsequent drop in that group's training accuracy is partly guaranteed by the selection rule rather than being an independent discovery about memorization. The claim that the identified neurons are the 'critical' memorization loci for that group is thus partially tautological for the gradient criterion.

full rationale

The paper's core descriptive finding—that a small set of neurons disproportionately affects minority-group training accuracy—is supported by two selection criteria, and the magnitude criterion is group-invariant, so the gradient-criterion circularity does not by itself make the finding an artifact. Tables 13–14 strengthen the memorization interpretation by showing that training accuracy is affected much more than test accuracy when those neurons are modified, and the random-initialization/noise controls provide additional context. The Stage II fine-tuning results (80.9% vs 64.0% WGA on Waterbirds/ResNet-50) are genuine improvements over ERM, but the paper does not run a no-pruning or random-mask control, so the improvement cannot be uniquely attributed to removing critical neurons; that is an experimental underdetermination rather than a derivation-by-construction circularity. There are no load-bearing self-citations, imported uniqueness theorems, or renamed known results that would force the central claim. Overall, the score reflects one partially self-referential evaluation step in the gradient-based tracing, while the independent magnitude-based evidence and external benchmarks keep the central claim from reducing to its inputs.

Assumptions & free parameters 6 free parameters · 5 assumptions · 2 invented entities

The central claims rest on domain assumptions about how memorization should be measured and about the validity of gradient/magnitude heuristics as memorization detectors. The framework introduces several tuned hyperparameters (pruning percentage, loss weight, kick-in epoch) whose values were selected by ablation to maximize worst-group accuracy, so the reported gains are conditional on those choices.

free parameters (6)

Pruning percentage (0.01%) = 0.01%
Chosen by ablation (Table 9); the framework's reported WGA gains depend on this value, and 10% gradient pruning collapses WGA to 0.04%.
Loss balancing term lambda = 0.2
Ablated in Table 8; lambda=0.2 gives the best WGA (80.89%), while lambda=0.01 gives 65.26% and lambda=10 gives 59.97%.
Kick-in epoch (40) = 40
ERM pretraining length before fine-tuning; Table 5 shows epoch 30 is ineffective, and 40/50/60 work, so the value is tuned to the point where ERM has converged.
Top-k pruned neurons in tracing = k=1,2,3
Used for the critical neuron evidence; ablation to 5 and 10 in Table 3 shows similar trends.
Hard-example batch size = 256 then 128
Used to compute gradients for neuron selection in the framework; not ablated.
Contrastive temperature tau = Not reported
Used in NT-Xent loss (Eq. 2.1); no value or ablation given in the paper.

assumptions (5)

domain assumption Memorization is operationalized as high training accuracy on atypical examples, not as literal storage of training inputs.
Defined in Section 1; the whole tracing analysis measures training accuracy changes, so the paper's notion of memorization rests on this equivalence.
domain assumption Gradient magnitude and weight magnitude identify the neurons responsible for memorizing a group.
Used throughout Section 2.1; no proof that these heuristics isolate memorization as opposed to other forms of importance.
domain assumption Pruning 0.01% of neurons does not substantially harm the model's general capabilities, so any accuracy drop is attributable to the removed information.
Required for interpreting the pruning experiments and the auxiliary model in Section 2.2.
domain assumption The Waterbirds and CelebA group labels correctly define the spurious correlation structure.
Group accuracy is computed against these labels; errors in annotation would misattribute effects.
domain assumption ImageNet-pretrained models are a valid starting point for studying memorization in this setting.
The high baseline accuracy of pretrained models is argued to be critical for memorization studies (Section 4.1).

invented entities (2)

Spurious memorization independent evidence
purpose: A postulated mechanism whereby atypical (minority) examples are memorized by a small subset of neurons during training, explaining the train-test accuracy gap.
Operationally defined and testable: pruning the identified neurons changes minority training accuracy, and the proposed fine-tuning claims to remove it; however, the causal evidence is incomplete due to missing controls.
Critical neurons independent evidence
purpose: The small subset of neurons claimed to carry spurious memorization; the target of pruning in the proposed framework.
Falsifiable handle: zeroing out these neurons (selected by gradient or magnitude) produces larger accuracy drops for minority groups than majority groups. The magnitude-based criterion is group-invariant, providing an external anchor.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Uncovering Memorization Effect in the Presence of Spurious Correlations." pith.science (2026). https://pith.science/paper/CXRWUJE4

@misc{pith2026250100961,
  author       = {Pith},
  title        = {Pith review of: Uncovering Memorization Effect in the Presence of Spurious Correlations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CXRWUJE4}},
  note         = {Machine review of arXiv:2501.00961}
}

read the original abstract

Machine learning models often rely on simple spurious features -- patterns in training data that correlate with targets but are not causally related to them, like image backgrounds in foreground classification. This reliance typically leads to imbalanced test performance across minority and majority groups. In this work, we take a closer look at the fundamental cause of such imbalanced performance through the lens of memorization, which refers to the ability to predict accurately on atypical examples (minority groups) in the training set but failing in achieving the same accuracy in the testing set. This paper systematically shows the ubiquitous existence of spurious features in a small set of neurons within the network, providing the first-ever evidence that memorization may contribute to imbalanced group performance. Through three experimental sources of converging empirical evidence, we find the property of a small subset of neurons or channels in memorizing minority group information. Inspired by these findings, we hypothesize that spurious memorization, concentrated within a small subset of neurons, plays a key role in driving imbalanced group performance. To further substantiate this hypothesis, we show that eliminating these unnecessary spurious memorization patterns via a novel framework during training can significantly affect the model performance on minority groups. Our experimental results across various architectures and benchmarks offer new insights on how neural networks encode core and spurious knowledge, laying the groundwork for future research in demystifying robustness to spurious correlation.

Figures

Figures reproduced from arXiv: 2501.00961 by the authors.

**Figure 1.** Imbalanced Group Performance on Waterbirds. Majority groups (G0 and G3) show the minimal gap between training and test accuracy, while minority groups (G1 and G2) yield a significantly larger discrepancy. Both models are trained with ERM. Here the star superscript (*) in the figure is used to emphasize the minority groups G1 and G2. The minority groups are atypical examples to neural networks (NNs), as these small… view at source ↗

**Figure 2.** Group accuracy change by pruning Top-k neuron(s) with gradient-based (left) and magnitudebased (right) criterion. Within each group G, three bars with gradated hues indicate the accuracy shift after zeroing out the neurons with the top-1, top-2, and top-3 largest gradients or magnitudes, respectively. Note that the minority groups G1 and G2 are emphasized with the star superscript (*). Our systematic study is perfo… view at source ↗

**Figure 3.** Largest gradient neurons’ magnitude ranking distribution (left) and largest magnitude neurons’ gradient ranking distribution (right). The x-axis percentage refers to the relative ranking of neurons’ gradients or magnitudes across the entire network (0%: smallest, 100%: largest). as f (mj ⊙θ,·), where mj is a mask with neurons in Ij being masked. The change in accuracy ∆acc for each group j is given by ∆acc(j) = |acc… view at source ↗

Figures from the paper (9 more)

**Figure 4.** Figure 4: Group accuracy change by random initialize Top-k neuron(s) with gradient-based (left) and magnitude-based (right) criterion. For each group, 3 bars with gradated hues indicate the accuracy shift after random initializing top-1, top-2, and top-3 neurons with the large…

**Figure 5.** Figure 5: Group accuracy change by adding random noise to Top-k neuron(s) with gradient-based (left) and magnitude-based (right) criterion. For each group, 3 bars of gradated hues indicate the group accuracy change after adding random noise to top-1, top-2, and top-3 neurons w…

**Figure 6.** Figure 6: Comparison of Worst Group Accuracy (WGA) across various architectures and datasets. The light-colored bars corresponds to the standard ERM training, and the dark color corresponds to our proposed framework. In this subsection, we perform extensive ablation studies to…

**Figure 7.** Figure 7: Group accuracy change by only prunning top-3 largest gradient or magnitude neurons within each conv layer in ResNet-50. The color bar is in the scale of percentage. 0 5 10 15 20 25 30 35 40 45 Layer Depth 0 * 1 * 2 Gradient-based 3 0 1 2 3 4 [PITH_FULL_IMAGE:figures/f…

**Figure 8.** Figure 8: Distribution of Critical Neurons in ResNet-50 via Unstructured Tracing. This analysis utilizes a gradient-based criterion to identify critical neurons, with the color bar indicating the neuron count. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

**Figure 9.** Figure 9: The overview of our proposed fine-tuning framework. The key innovation of our framework lies in its dual-branch architecture, consisting of the target model and an auxiliary model, the latter being a pruned version of the target with critical neurons removed The weight…

**Figure 10.** Figure 10: GradCAM visualization of Waterbirds on ResNet-50 by ERM (top two rows) and our fine-tuning [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]

**Figure 11.** Figure 11: Group accuracy change by random initialize Top-k neuron(s) with the heuristics of gradient (top row) and magnitude (bottom row). Within each group, three bars with gradated hues indicate the accuracy shift after random initializing the top-1, top-2, and top-3 neurons …

**Figure 12.** Figure 12: Group accuracy change by adding random noise to Top-k neuron(s) with the heuristics of gradient (top row) and magnitude (bottom row). Within each group, three bars with gradated hues indicate the accuracy shift after random initializing top-1, top-2, and top-3 neurons…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 13 canonical work pages

[1]

Deep learning through the lens of example difficulty

Robert Baldock, Hartmut Maennel, and Behnam Neyshabur. Deep learning through the lens of example difficulty. In Advances in Neural Information Processing Systems, volume 34, pages 10876--10889, 2021

work page 2021
[2]

Recognition in terra incognita

Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. In Proceedings of the European conference on computer vision (ECCV), pages 456--473, 2018

work page 2018
[3]

Demographic dialectal variation in social media: A case study of african-american english

Su Lin Blodgett, Lisa Green, and Brendan O’Connor. Demographic dialectal variation in social media: A case study of african-american english. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1119--1130, 2016

work page 2016
[4]

Gender shades: Intersectional accuracy disparities in commercial gender classification

Joy Buolamwini and Timnit Gebru. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency, pages 77--91. PMLR, 2018

2018
[5]

A simple framework for contrastive learning of visual representations

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. In International conference on machine learning, pages 1597--1607. PMLR, 2020

2020
[6]

Imagenet: A large-scale hierarchical image database

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In IEEE conference on computer vision and pattern recognition, pages 248--255. IEEE, 2009

work page 2009
[7]

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

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations, 2021. URL https...

2021
[8]

Shortcut learning in deep neural networks

Robert Geirhos, J \"o rn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2 0 (11): 0 665--673, 2020

2020

Show all 24 references

[9]

Fairness without demographics in repeated loss minimization

Tatsunori Hashimoto, Megha Srivastava, Hongseok Namkoong, and Percy Liang. Fairness without demographics in repeated loss minimization. In International Conference on Machine Learning, pages 1929--1938. PMLR, 2018

1929
[10]

Deep residual learning for image recognition

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770--778, 2016

2016
[11]

What do compressed deep neural networks forget? arXiv preprint arXiv:1911.05248, 2019

Sara Hooker, Aaron Courville, Gregory Clark, Yann Dauphin, and Andrea Frome. What do compressed deep neural networks forget? arXiv preprint arXiv:1911.05248, 2019

1911 arXiv
[12]

Evaluation of neural architectures trained with square loss vs cross-entropy in classification tasks

Like Hui and Mikhail Belkin. Evaluation of neural architectures trained with square loss vs cross-entropy in classification tasks. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=hsFN92eQEla

2021
[13]

Simple data balancing achieves competitive worst-group-accuracy

Badr Youbi Idrissi, Martin Arjovsky, Mohammad Pezeshki, and David Lopez-Paz. Simple data balancing achieves competitive worst-group-accuracy. In Conference on Causal Learning and Reasoning, pages 336--351. PMLR, 2022

2022
[14]

Last layer re-training is sufficient for robustness to spurious correlations

Polina Kirichenko, Pavel Izmailov, and Andrew Gordon Wilson. Last layer re-training is sufficient for robustness to spurious correlations. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=Zb6c8A-Fghk

2023
[15]

Deep learning face attributes in the wild

Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In Proceedings of the IEEE international conference on computer vision, pages 3730--3738, 2015

2015
[16]

Can neural network memorization be localized? In Proceedings of the 40th International Conference on Machine Learning, volume 202, pages 23536--23557

Pratyush Maini, Michael C Mozer, Hanie Sedghi, Zachary C Lipton, J Zico Kolter, and Chiyuan Zhang. Can neural network memorization be localized? In Proceedings of the 40th International Conference on Machine Learning, volume 202, pages 23536--23557. PMLR, 2023

2023
[17]

Simple and fast group robustness by automatic feature reweighting

Shikai Qiu, Andres Potapczynski, Pavel Izmailov, and Andrew Gordon Wilson. Simple and fast group robustness by automatic feature reweighting. In International Conference on Machine Learning, pages 28448--28467. PMLR, 2023

2023
[18]

Learning transferable visual models from natural language supervision

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

2021
[19]

``why should i trust you?'' explaining the predictions of any classifier

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. ``why should i trust you?'' explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 1135--1144, 2016

2016
[20]

Hashimoto, and Percy Liang

Shiori Sagawa, Pang Wei Koh, Tatsunori B. Hashimoto, and Percy Liang. Distributionally robust neural networks. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id=ryxGuJrFvS

2020
[21]

Improved deep metric learning with multi-class n-pair loss objective

Kihyuk Sohn. Improved deep metric learning with multi-class n-pair loss objective. Advances in neural information processing systems, 29, 2016

2016
[22]

On the geometry of generalization and memorization in deep neural networks

Cory Stephenson, suchismita padhy, Abhinav Ganesh, Yue Hui, Hanlin Tang, and SueYeon Chung. On the geometry of generalization and memorization in deep neural networks. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=V8jrrnwGbuc

2021
[23]

C. Wah, S. Branson, P. Welinder, P. Perona, and S. Belongie. The caltech-ucsd birds-200-2011 dataset. Technical Report CNS-TR-2011-001, California Institute of Technology, Computation & Neural Systems Technical Report, 2011

2011
[24]

Mitigating spurious correlations in multi-modal models during fine-tuning

Yu Yang, Besmira Nushi, Hamid Palangi, and Baharan Mirzasoleiman. Mitigating spurious correlations in multi-modal models during fine-tuning. In International Conference on Machine Learning, pages 39365--39379. PMLR, 2023

2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Deep learning through the lens of example difficulty

Robert Baldock, Hartmut Maennel, and Behnam Neyshabur. Deep learning through the lens of example difficulty. In Advances in Neural Information Processing Systems, volume 34, pages 10876--10889, 2021

work page 2021

[2] [2]

Recognition in terra incognita

Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. In Proceedings of the European conference on computer vision (ECCV), pages 456--473, 2018

work page 2018

[3] [3]

Demographic dialectal variation in social media: A case study of african-american english

Su Lin Blodgett, Lisa Green, and Brendan O’Connor. Demographic dialectal variation in social media: A case study of african-american english. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1119--1130, 2016

work page 2016

[4] [4]

Gender shades: Intersectional accuracy disparities in commercial gender classification

Joy Buolamwini and Timnit Gebru. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency, pages 77--91. PMLR, 2018

2018

[5] [5]

A simple framework for contrastive learning of visual representations

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. In International conference on machine learning, pages 1597--1607. PMLR, 2020

2020

[6] [6]

Imagenet: A large-scale hierarchical image database

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In IEEE conference on computer vision and pattern recognition, pages 248--255. IEEE, 2009

work page 2009

[7] [7]

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

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations, 2021. URL https...

2021

[8] [8]

Shortcut learning in deep neural networks

Robert Geirhos, J \"o rn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2 0 (11): 0 665--673, 2020

2020