Knowledge Swapping via Learning and Unlearning

Lechao Cheng; Meng Wang; Mingyu Xing; Shengeng Tang; Yaxiong Wang; Zhun Zhong

REVIEW 3 major objections 6 minor 16 references

Knowledge Swapping via Learning and Unlearning

T0 review · 3 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read This paper introduces Knowledge Swapping, a task that forgets specified classes, keeps the rest, and learns new ones, and argues the reliable recipe is to learn first and forget second.

desk verdict A useful new task and benchmark, but the claimed 'learning before forgetting' principle is confounded with which phase is last, and the feature-hierarchy story is not supported by the paper's own control. read the letter →

arxiv 2502.08075 v2 pith:V2MAJTJM submitted 2025-02-12 cs.CV

Mingyu Xing , Lechao Cheng , Shengeng Tang , Yaxiong Wang , Zhun Zhong , Meng Wang This is my paper · ORCID

classification cs.CV

keywords KnowledgeSwappingmachineunlearningcontinuallearningBeforeForgettingfeaturehierarchyLoRAgroupsparseregularizationselective

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 Knowledge Swapping, a task in which a pretrained model must forget user-specified knowledge, keep essential knowledge, and absorb new knowledge in one update. It claims the order of the two phases decides success: train the new task first, then erase the forget-set, rather than the intuitive forget-then-learn. The reason offered is a directional feature hierarchy: incremental learning builds from low-level to high-level features, while forgetting travels from high-level semantics down to low-level features. If the claim holds, any knowledge-editing pipeline can be made more reliable simply by sequencing its fine-tuning and unlearning stages, with forget-set accuracy dropping to near zero while retention and new-task accuracy stay high.

What carries the argument

The load-bearing object is the directional feature hierarchy, operationalized through low-rank adaptation matrices $\Delta W = AB$ attached to the FFN linear layers of each transformer block, with group sparse regularization on $\lVert A_k \rVert_F^2 + \lVert B_k \rVert_F^2$ to retain only needed modules. The forgetting phase uses a boundary-constrained loss $\mathcal{L}_{\text{forget}} = \mathrm{ReLU}(\mathrm{BND} - \mathcal{L}(f(X_f),Y_f))$ so erasure is stabilized while retention and learning losses continue. The machinery converts the ordering insight into a concrete two-stage update: learn low-level features before high-level erasure, so the forgetting pass stays confined to semantic layers and cannot be undone by later low-level changes.

What would settle it

Measure per-block parameter-change statistics during learn-only and forget-only runs on the classification and detection models used in the paper's main tables: if learning updates are not concentrated in early blocks or forgetting updates are not concentrated in late blocks, the directional hierarchy that motivates the learn-first order is falsified.

Watch

Extended reading notes

Core claim

The central discovery is the directional feature hierarchy: in the studied transformer models, incremental learning progresses from low-level representations to higher-level semantic features, whereas targeted forgetting moves from high-level semantics down to low-level features. From this the paper derives the Learning Before Forgetting principle, a two-stage procedure that first fine-tunes on the learning set (with retention-set supervision) and only then runs selective forgetting. Empirically, this single ordering choice drives forget-set accuracy, mIoU, or mAP to near zero across image classification, object detection, and semantic segmentation, while the reverse order lets forgotten classes rebound after the learning phase because low-level changes invalidate the earlier high-level erasure.

Load-bearing premise

The entire ordering argument rests on the claim that learning always builds from low-level to high-level features while forgetting erases from high-level down to low-level, a directionality the paper reads off parameter-norm plots rather than proves with quantitative layer-wise statistics.

Editorial extensions

If this is right

For any two-stage knowledge-editing pipeline on a pretrained transformer, the paper predicts learn-first beats forget-first: with low-rank adaptation and group sparsity, forget-set accuracy falls to near zero while retention and learning accuracy stay high.
The reverse order should show a rebound: after an initial drop, forgetting-set accuracy climbs back because the learning phase perturbs low-level parameters that the earlier forgetting pass relied on.
Adding a second forgetting pass after forget-then-learn restores erasure, which the paper reads as evidence that the rebound is caused by low-level re-activation rather than incomplete erasure.
The ordering advantage transfers across tasks and architectures: image classification on ImageNet-100, semantic segmentation on ADE20K, and object detection on COCO all follow the same learn-first pattern.

Reading between the lines

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

If the directional hierarchy is a general property of transformer training, the learn-then-forget ordering should extend beyond low-rank adaptation to other parameter-efficient editing schemes, including unlearning in large language models: fine-tune on new capabilities before erasing unwanted behaviors.
The paper's evidence for hierarchy direction is visual inspection of L2 norm differences on three segmentation settings; a quantitative per-block change statistic with variance bars would turn the principle into a measurable invariant and could predict which classes are easy or hard to forget.
The boundary loss suggests a testable extension: sweeping the BND threshold and measuring the forget-versus-retain trade-off curve would reveal whether near-zero forgetting accuracy is a robust plateau or a knife-edge result.

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.

Desk Editor's Note

read the letter

Colleague,

The paper on Knowledge Swapping is worth a look. It proposes a new task—simultaneously learn new classes, forget specified classes, and retain the rest on a pretrained model—and backs it with consistent experiments across classification, segmentation, and detection. The main empirical result, that learn-then-forget beats forget-then-learn, is clear-cut and large in magnitude. That alone is a contribution: the task formulation and benchmark are genuinely new and the code is promised.

The problem is the paper's interpretation. The authors argue that learning goes from low-level to high-level features while forgetting goes high-to-low, and therefore learning first is better. But their own Table 2 contains a control that quietly undermines this. In the F→L→F row, they add a final forgetting phase after the forget-then-learn sequence, and the forgetting mIoU drops back to near zero. That means the initial order barely matters; what matters is that the last phase is forgetting. The L→F advantage is exactly what you'd expect if the objective of the final phase directly drives the forgetting metric. In F→L the final phase is learning, and the learning loss has no forgetting term, so forgetting accuracy rebounds. That is a confound, not evidence for a feature hierarchy.

The feature-hierarchy "Discovery-I" is itself weak: it comes from visual inspection of L2-norm curves, with no layer-wise statistics, no error bars, and no check on the classification or detection models. The paper also doesn't compare against any existing continual learning or unlearning method, reports single runs without seeds, and tunes hyperparameters per task without sensitivity analysis.

None of this kills the practical value of the task or the benchmark. A practitioner who wants to swap knowledge on a pretrained model can follow "learn then forget" and get good numbers—that's useful. But the mechanistic story should be dropped or heavily qualified. The paper needs a major revision that compares L→F against F→L→F in every setting, reports variance, and tests the hierarchy quantitatively. It deserves a serious referee, but not as is.

I'd bring it to a reading group to discuss the confound, and I'd probably cite the task definition in future work on unlearning. Send it to review with a strong request for revision.

Referee Report

3 major / 6 minor

Summary. The paper introduces Knowledge Swapping, a task in which a pretrained model must simultaneously learn new knowledge, retain existing knowledge, and forget user-specified knowledge. The authors propose a two-stage pipeline built on LoRA and group sparse regularization, and claim that performing the learning phase before the forgetting phase ('Learning Before Forgetting') yields substantially better results than the reverse order. The claimed explanation is a feature-hierarchy discovery: incremental learning progresses from low-level to high-level features, while forgetting proceeds from high-level to low-level features. The paper validates the strategy with experiments on ImageNet-100 classification, COCO-based object detection, and ADE20K-based semantic segmentation, and releases source code.

Significance. If the central claim is correct, the paper offers a simple and practically useful design principle for selective model editing: when swapping knowledge, train the new task before unlearning the old one. The task formulation itself is clear and the experimental sweep across three vision tasks is a strength, as is the public code release. However, the paper's key non-tautological claim—that the order matters because of a directional feature hierarchy—is currently undermined by a comparison that confounds order with which objective is optimized last, and by evidence for the hierarchy that is only qualitative. The benchmark is therefore not yet established to the standard needed for a strong claim.

major comments (3)

[§5.3, Table 2] The central comparison L→F versus F→L is confounded because the two sequences end with different optimization objectives. In L→F, the final phase minimizes Eq. (8), which directly drives the forgetting-set metric toward zero; in F→L, the final phase minimizes Eq. (7), which contains no forgetting term. The rebound of mIoUf after F→L is therefore expected even without any feature-hierarchy effect. This is confirmed by the F→L→F rows of Table 2, where adding a single final forgetting phase reduces mIoUf to 0.08–0.39 across the four datasets, comparable to the L→F rows. The large gap in Tables 1–3 between L→F and F→L thus does not establish the Learning Before Forgetting principle; it is consistent with the simpler rule that the unlearning phase should be last. Please add conditions that hold the final phase fixed while varying the order, or otherwise isolate the order effect.
[§3.2, Discovery-I] Discovery-I is the sole mechanistic basis for the Learning Before Forgetting principle, but it is inferred from visual inspection of L2-norm and gradient curves (Figures 2, 4, and 6) on three semantic-segmentation settings. There are no quantitative layer-wise statistics, no error bars, and no analogous measurement on the classification or detection models used in Tables 1 and 3. The claim that learning progresses from low-level to high-level features and forgetting from high-level to low-level is therefore not tested on the models where the main results are reported. Please provide per-layer statistics (e.g., mean and variance of weight updates relative to pretrained weights) for all three task families, and check whether the observed pattern could be an artifact of the group-sparse LoRA regularization rather than an inherent property of learning and forgetting.
[§5.1–§5.4, Tables 1–3] All results are single runs with no seeds or error bars. For a benchmark paper, variance estimates across at least three seeds are necessary to assess whether the observed margins are repeatable and whether the per-task hyperparameter choices (α, β, and BND reported in §5.1) materially affect the conclusions. Without such estimates, the quantitative claims in Sections 5.2–5.4 are not yet established to the standard expected for a benchmark.

minor comments (6)

[§4.3] The heading 'Sparse Constriant' should be 'Sparse Constraint'.
[§2.2] The word 'whiestimatetes' in the related-work paragraph should be 'which estimates'.
[Tables 1–3] The notation in the tables, e.g., 'L− →F', is hard to parse; please use a consistent arrow notation and explain it in the captions.
[§3.2] The term 'knock-on feature hierarchy' is non-standard; consider renaming it to something like 'feature-hierarchy interplay' for clarity.
[§5.1] Oxford-IIIT Pet appears as a learning set in both the classification and segmentation setups; please clarify whether the splits and data composition differ between the two tasks.
[§2 and §5] The paper surveys many continual-learning and machine-unlearning methods but includes none of them as experimental baselines; adding at least one adapted baseline would make the proposed benchmark more informative.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the order comparison is empirical and the reported scores are measured outcomes, not fitted parameters or definitional identities.

full rationale

The paper's central claim is an empirical ordering strategy (Learning Before Forgetting). The losses in Eqs. 5-9 define two distinct pipelines; L→F and F→L are different optimization sequences, and the reported accuracy/mIoU values are measured results, not quantities that reduce to the loss definitions by construction. Discovery-I (Section 3.2) is inferred from the same segmentation experiments later used in Table 2, which is an in-sample-evidence limitation and a potential confound: F→L ends with a learning phase whose loss (Eqs. 5-7) has no forgetting term, so the rebound of forgetting accuracy is partly expected; the F→L→F rows in Table 2 further suggest the final phase dominates. However, this is a confound or alternative explanation, not a circular derivation: no equation is defined in terms of the conclusion, no fitted parameter is renamed as a prediction, and no load-bearing self-citation is used. The method relies on external, independently established components (LoRA, Lasso/group sparsity, standard losses). Hence no formal circularity; the concern belongs to correctness or experimental design rather than circularity.

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

No new physical entities are introduced. The method relies on explicit free hyperparameters (alpha, beta, BND) and on unproven structural assumptions about where knowledge lives in a transformer and how layer semantics are ordered.

free parameters (4)

alpha (learning and forgetting phases) = 0.05 for classification learning; 0.01 for detection/segmentation learning and all forgetting phases
Regularization weight on the group-sparsity term L_re in Eqs. 7 and 9. Set per task with no sensitivity analysis.
beta (learning phase) = 0.2 (classification), 0.9 (detection/segmentation)
Balances L_learn against L_retain in Eq. 7. Set per task, no sensitivity analysis.
beta (forgetting phase) = 0.2 (all tasks)
Balances the bounded forgetting loss L_forget in Eq. 9. Set per task, no sensitivity analysis.
BND (forgetting boundary) = 105 (classification), 15 (detection), 115 (segmentation)
Threshold in ReLU(BND - L_forget) to stabilize the unlearning gradient. Chosen per task without ablation.

assumptions (5)

domain assumption Transformer linear layers, particularly FFNs, are the primary storage of model knowledge (citing Geva et al., 2020).
Used in Section 4.2 to justify updating only LoRA adapters on linear layers. If false, the sparse unlearning may miss knowledge stored elsewhere.
domain assumption Earlier layers encode low-level features and later layers encode high-level semantics in the tested models.
Underpins Discovery-I and the Learning Before Forgetting principle in Section 3.2. Not verified for ViT, Mask2Former, or DINO beyond visual inspection of weight norms.
standard math Group Lasso regularization with Frobenius norms induces group sparsity.
Standard property used in Eq. 4, not derived in the paper.
domain assumption Group Lasso on LoRA A and B matrices selectively isolates task-relevant modules without damaging retained knowledge.
Assumed in Section 4.3. No theorem or ablation isolates the contribution of L_re to the ordering result.
domain assumption The forgetting and retention sets are disjoint subsets of pretraining data and the learning set is a new task.
Task definition in Section 3.1 requires this data availability. In practice, access to subsets of the pretraining data may not be granted.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Knowledge Swapping via Learning and Unlearning." pith.science (2026). https://pith.science/paper/V2MAJTJM

@misc{pith2026250208075,
  author       = {Pith},
  title        = {Pith review of: Knowledge Swapping via Learning and Unlearning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V2MAJTJM}},
  note         = {Machine review of arXiv:2502.08075}
}

read the original abstract

We introduce \textbf{Knowledge Swapping}, a novel task designed to selectively regulate knowledge of a pretrained model by enabling the forgetting of user\-specified information, retaining essential knowledge, and acquiring new knowledge simultaneously. By delving into the analysis of knock-on feature hierarchy, we find that incremental learning typically progresses from low\-level representations to higher\-level semantics, whereas forgetting tends to occur in the opposite direction\-starting from high-level semantics and moving down to low-level features. Building upon this, we propose to benchmark the knowledge swapping task with the strategy of \textit{Learning Before Forgetting}. Comprehensive experiments on various tasks like image classification, object detection, and semantic segmentation validate the effectiveness of the proposed strategy. The source code is available at \href{https://github.com/xingmingyu123456/KnowledgeSwapping}{https://github.com/xingmingyu123456/KnowledgeSwapping}.

Figures

Figures reproduced from arXiv: 2502.08075 by the authors.

**Figure 1.** Comparison of three tasks: Continuous Learning, Machine Unlearning, and our Knowledge Swapping. lateral forgetting of specific content within pretrained models. Nonetheless, approaches that simultaneously enable the learning of new knowledge and the forgetting of specific content remain underexplored. Inspired by this insight, we propose a novel task termed Knowledge Swapping, which enables the selective forgetti… view at source ↗

**Figure 2.** L2 norm for each parameter under L → F and F → L. The superscript W denotes the weight norm value at the current stage. The figure illustrates that (a) during the Learning Before Forgetting phase, changes in parameter norms are predominantly concentrated in layers responsible for high-level semantic representations. Conversely, (b) in the Learning After Forgetting phase, parameter norm changes primarily occur in lay… view at source ↗

**Figure 3.** Benchmark Framework. First, we decouple knowledge swapping into separate learning and forgetting processes. We observed that the learning process progresses from low-level features to high-level features, while the forgetting process proceeds in the opposite direction—from high-level features to low-level features. Therefore, a two-stage strategy of Learning Before Forgetting is adopted. In general, we adopt LoRA to… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Logarithm of the Average Gradient. We compute the logarithm of cumulative average gradient changes at different stages in the L → F and F → L processes. We observe two key phenomena: first, parameter changes during the learning phases (L G → F and F → L G) are consiste…

**Figure 5.** Figure 5: Qualitative results on semantic segmentation. The forgotten classes are marked with red dotted lines, and the learned class is marked with dark green dotted lines. procedure VOC Oxford-pet COCO Deepglobe Land mIoUr ↑ mIoUl ↑ mIoUf ↓ mIoUr ↑ mIoUl ↑ mIoUf ↓ mIoUr ↑ mIoU…

**Figure 6.** Figure 6: L2 norm of weights in F → L → F viously tuned high-level forgetting parameters ineffective [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: highlights this issue, showing that mountain, which is initially erased and blended into sand, re-emerges after learning, demonstrating the instability of this approach. 5.4. Object Detection Results procedure Cub Oxford-dog mAPr ↑ mAPl ↑ mAPf ↓ mAPr ↑ mAPl ↑ mAPf ↓ St…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 2 canonical work pages

[4]

K., and Hays, J

Hoiem, D., Divvala, S. K., and Hays, J. H. Pascal voc 2008 challenge. World Literature Today, 24(1):1–4,

work page 2008
[10]

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
[13]

The caltech-ucsd birds-200-2011 dataset

Wah, C., Branson, S., Welinder, P., Perona, P., and Belongie, S. The caltech-ucsd birds-200-2011 dataset

2011
[15]

Defensive unlearning with adversarial training for robust concept erasure in diffusion models

Zhang, Y ., Chen, X., Jia, J., Zhang, Y ., Fan, C., Liu, J., Hong, M., Ding, K., and Liu, S. Defensive unlearning with adversarial training for robust concept erasure in diffusion models. arXiv preprint arXiv:2405.15234,

arXiv
[16]

A model or 603 exemplars: Towards memory-efficient class- incremental learning

Zhou, D.-W., Wang, Q.-W., Ye, H.-J., and Zhan, D.-C. A model or 603 exemplars: Towards memory-efficient class- incremental learning. arXiv preprint arXiv:2205.13218,

arXiv
[2006]

M., and Shum, H.-Y

Zhang, H., Li, F., Liu, S., Zhang, L., Su, H., Zhu, J., Ni, L. M., and Shum, H.-Y . Dino: Detr with improved denois- ing anchor boxes for end-to-end object detection. arXiv preprint arXiv:2203.03605,

arXiv
[2009]

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,

arXiv
[2011]

Deep- globe 2018: A challenge to parse the earth through satel- lite images

Demir, I., Koperski, K., Lindenbaum, D., Pang, G., Huang, J., Basu, S., Hughes, F., Tuia, D., and Raskar, R. Deep- globe 2018: A challenge to parse the earth through satel- lite images. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops, pp. 172–181,

work page 2018

Show all 16 references

[2012]

Rebuffi, S.-A., Kolesnikov, A., Sperl, G., and Lampert, C. H. icarl: Incremental classifier and representation learning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pp. 2001–2010,

2001
[2015]

and Liu, H

Liu, H. and Liu, H. Continual learning with recursive gra- dient optimization. arXiv preprint arXiv:2201.12522 ,

arXiv
[2017]

A., Rabinowitz, N

Rusu, A. A., Rabinowitz, N. C., Desjardins, G., Soyer, H., Kirkpatrick, J., Kavukcuoglu, K., Pascanu, R., and Had- sell, R. Progressive neural networks. arXiv preprint arXiv:1606.04671,

arXiv
[2019]

Transformer feed-forward layers are key-value memories

Geva, M., Schuster, R., Berant, J., and Levy, O. Transformer feed-forward layers are key-value memories. arXiv preprint arXiv:2012.14913,

2012 arXiv
[2021]

Llm-assisted multi-teacher continual learning for visual question answering in robotic surgery

Chen, K., Du, Y ., You, T., Islam, M., Guo, Z., Jin, Y ., Chen, G., and Heng, P.-A. Llm-assisted multi-teacher continual learning for visual question answering in robotic surgery. arXiv preprint arXiv:2402.16664,

arXiv
[2022]

Model sparsity can simplify machine unlearning

9 Knowledge Swapping via Learning and Unlearning Liu, J., Ram, P., Yao, Y ., Liu, G., Liu, Y ., SHARMA, P., Liu, S., et al. Model sparsity can simplify machine unlearning. Advances in Neural Information Processing Systems, 36, 2024a. Liu, S., Yao, Y ., Jia, J., Casper, S., Bar...

arXiv
[2023]

Lin, T.-Y ., Maire, M., Belongie, S., Hays, J., Perona, P., Ra- manan, D., Doll´ar, P., and Zitnick, C. L. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, pp. 740–

2014
[2024]

Atlas: Adapter-based multi-modal continual learning with a two-stage learning strategy

Li, H., Tan, Z., Li, X., and Huang, W. Atlas: Adapter-based multi-modal continual learning with a two-stage learning strategy. arXiv preprint arXiv:2410.10923,

arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [4]

K., and Hays, J

Hoiem, D., Divvala, S. K., and Hays, J. H. Pascal voc 2008 challenge. World Literature Today, 24(1):1–4,

work page 2008

[2] [10]

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

[3] [13]

The caltech-ucsd birds-200-2011 dataset

Wah, C., Branson, S., Welinder, P., Perona, P., and Belongie, S. The caltech-ucsd birds-200-2011 dataset

2011

[4] [15]

Defensive unlearning with adversarial training for robust concept erasure in diffusion models

Zhang, Y ., Chen, X., Jia, J., Zhang, Y ., Fan, C., Liu, J., Hong, M., Ding, K., and Liu, S. Defensive unlearning with adversarial training for robust concept erasure in diffusion models. arXiv preprint arXiv:2405.15234,

arXiv

[5] [16]

A model or 603 exemplars: Towards memory-efficient class- incremental learning

Zhou, D.-W., Wang, Q.-W., Ye, H.-J., and Zhan, D.-C. A model or 603 exemplars: Towards memory-efficient class- incremental learning. arXiv preprint arXiv:2205.13218,

arXiv

[6] [2006]

M., and Shum, H.-Y

Zhang, H., Li, F., Liu, S., Zhang, L., Su, H., Zhu, J., Ni, L. M., and Shum, H.-Y . Dino: Detr with improved denois- ing anchor boxes for end-to-end object detection. arXiv preprint arXiv:2203.03605,

arXiv

[7] [2009]

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,

arXiv

[8] [2011]

Deep- globe 2018: A challenge to parse the earth through satel- lite images

Demir, I., Koperski, K., Lindenbaum, D., Pang, G., Huang, J., Basu, S., Hughes, F., Tuia, D., and Raskar, R. Deep- globe 2018: A challenge to parse the earth through satel- lite images. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops, pp. 172–181,

work page 2018