Pith. sign in

REVIEW 4 major objections 5 minor 119 references

Transforming Vision Transformer: Towards Efficient Multi-Task Asynchronous Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Training multiple vision tasks in one ViT can outdo separate fine-tuning while tuning only about one million parameters and adding no inference cost.

desk verdict A sensible training-time MTL recipe with plausible but statistically unsupported gains; deserves peer review, but the error-bar claim and test-set hyperparameter picking need fixing. read the letter →

arxiv 2501.06884 v1 pith:NUPMWV3M submitted 2025-01-12 cs.CV

Hanwen Zhong , Jiaxin Chen , Yutong Zhang , Di Huang , Yunhong Wang This is my paper · ORCID
classification cs.CV
keywords multi-tasklearningvisiontransformermixtureofexpertslow-rankadaptationparameter-efficientfine-tuningasynchronouslogitdistillationrouterfading
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 claims that a pre-trained Vision Transformer can be turned into an efficient multi-task learner by decomposing its feed-forward layers into low-rank mixture-of-experts, fine-tuning those experts with LoRA, and then re-fusing everything back into the original architecture. It argues that forcing tasks to converge synchronously is the wrong instinct; instead, tasks should converge at their own pace while the logits of already-converged tasks are preserved. If correct, EMTAL provides a practical recipe: one unified model that beats separate full fine-tuning on several fine-grained classification benchmarks by 3-4% top-1 accuracy, using only 0.75-1.20M tunable parameters, and no slower at inference because the router fades away. The same recipe also improves VTAB-1k classification and NYUv2 dense prediction.

What carries the argument

The machinery is a triad of named components. MoEfied LoRA decomposes each FFN into a Mixture of Low-rank Experts by balanced $k$-means clustering of similar columns of $W_{up}$ (and corresponding rows of $W_{down}$), then applies LoRA to each expert. Quality Retaining stores an exponential-moving-average logit bank $Z$ per class and adds a distillation loss weighted by $1/L_{\mathrm{CE},t}$ so converged tasks keep their knowledge. Router fading linearly blends the router output toward a constant, eventually setting $\alpha=0$, which lets the trained experts and LoRA updates be reparameterized back into the original FFN weights for zero added inference cost.

What would settle it

A direct test: run EMTAL with the QR term intact but replace the EMA logits with random or deliberately degraded logits, or set $m$ close to 1 so the bank stores early-training logits; if the mean accuracy on Multi-task FGVC barely moves, the QR mechanism is not doing the work attributed to it. A cheaper check is to sweep the unreported momentum $m$ and see whether the gap between MoEfied LoRA alone (90.27) and full EMTAL-4 (91.73) persists across values of $m$.

Watch

Extended reading notes

Core claim

The central discovery is that the rigid coupling of MoE and LoRA in earlier multi-task learning can be replaced by MoEfied LoRA: cluster the columns of the FFN up-projection, and matching rows of the down-projection, into $K$ groups of similar channels, treat each group as a low-rank expert, and tune each expert with LoRA. Because similar-weight columns form low-rank experts, LoRA's low-rank reparameterization becomes natural rather than forced. During training, a sample-driven soft router mixes the experts; a Quality Retaining loss, computed as a KL divergence toward an EMA of per-class logits weighted by the inverse of each task's cross-entropy loss, stops well-trained tasks from degrading while harder tasks keep optimizing. A router fading schedule then reduces the router's weight $\alpha$ to zero, allowing the learned LoRA deltas and composed expert matrices to be merged back into the original weight matrices, producing a unified static model with no routing overhead at inference.

Load-bearing premise

The load-bearing premise is that the EMA logit bank $Z$ in Eq. (10) stores genuinely high-quality predictions for already-converged tasks, so pulling current logits toward it with weight $1/L_{\mathrm{CE},t}$ preserves rather than distorts performance; the paper does not report the momentum $m$ or validate the bank's quality.

Editorial extensions

If this is right

  • Multi-task fine-tuning no longer requires a routing network at inference time, because the learned experts and LoRA deltas merge into the original FFN weights and the inference time stays at 7.15 ms, equal to the base ViT.
  • Letting tasks converge asynchronously and distilling from the EMA logit bank yields higher accuracy than gradient-balancing and loss-balancing multi-task optimization methods on Multi-task FGVC, VTAB-1k, and NYUv2.
  • Parameter cost drops to 0.75-1.20M tunable parameters versus 343.92M for separate full fine-tuning, so large ViTs can be multi-task adapted on a single GPU.
  • Accuracy improves with expert rank: EMTAL-4 reaches 91.73 mean top-1 on Multi-task FGVC, surpassing Aligned-MTL's 90.17 with 2.82M parameters at the same inference time.
  • The method also helps in few-shot settings; at 16 shots it improves over the second-best method by about 5% on the Multi-task FGVC benchmark.

Reading between the lines

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

  • An extension the paper leaves implicit is that the same decompose-tune-refuse recipe could apply to other transformer components, such as attention projections or the patch embedding, potentially widening the parameter-efficiency gains.
  • Because the QR loss is a form of self-distillation, its benefit should depend on the quality of the EMA bank; a testable variant would replace the EMA with a delayed copy of the model or a checkpointed teacher, which would clarify whether the mechanism is preserving knowledge or merely regularizing.
  • The paper observes that different tasks and different layers prefer different expert ranks, so an adaptive rank-allocation scheme is a natural next step that may exceed the fixed-rank EMTAL-4 results reported here.
  • QR targets temporal forgetting while gradient-based methods target instantaneous conflicts, so combining QR with a gradient-balancing method may be an untested but plausible way to gain further multi-task accuracy.
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 / 5 minor

Summary. The paper proposes EMTAL, a parameter-efficient multi-task learning method for Vision Transformers. The approach has three components: (i) MoEfied LoRA, which decomposes each FFN into low-rank experts by balanced k-means clustering of the up-projection weights and applies LoRA to each expert; (ii) a Quality Retaining (QR) optimization mechanism that distills current logits toward an EMA class-logit bank, weighted by the inverse per-task cross-entropy loss, to preserve already-converged tasks; and (iii) a router-fading strategy that anneals the router weight to zero so that the learned LoRA and expert structure can be reparameterized back into the original FFN, yielding no inference overhead. The paper reports consistent accuracy gains on Multi-task FGVC over Separate FT and state-of-the-art MTL baselines, gains on VTAB-1k and NYUv2, and claims a unified model with 0.75--1.20M tunable parameters and unchanged inference latency.

Significance. If the empirical claims hold, EMTAL is a practically valuable contribution: it combines MoE-style diversity with LoRA parameter efficiency and avoids the inference cost of dynamic routing via reparameterization. The reported results are potentially state-of-the-art for parameter-efficient MTL (e.g., 91.73 mean accuracy with 1.20M tunable params on Multi-task FGVC versus 90.17 for Aligned-MTL). The paper has genuine strengths: component-level ablations, a comparison of different expert-construction strategies, a low-rank analysis using the Ky Fan 2-k norm, and experiments across multiple backbones, few-shot settings, and dense-prediction tasks. However, the central empirical claim is currently supported only by single-run accuracies, and the largest measured gain comes from the QR mechanism, whose EMA logit bank and inverse-loss weighting are not validated. The significance is therefore conditional on additional statistical and mechanism-level evidence.

major comments (4)
  1. [§3.5 / Checklist item 7] The checklist item 7 states that error bars are described in Section 3.5, but Section 3.5 contains only low-rank property plots (Figure 4) and no accuracy error bars; all accuracy tables (Tables 1, 2, 3, 4, 6, 7) report single runs without variance or significance tests. Since the main claims rest on margins of 1--4 percentage points (e.g., EMTAL-4 91.73 vs. Aligned-MTL 90.17 in Table 1; EMTAL-4 87.89 vs. MOELoRA 87.01 in Table 2), the paper needs multiple seeds with error bars or confidence intervals for at least the main benchmark comparisons before the improvements can be assessed.
  2. [§3.1, Table 8, Table 6] Multi-task FGVC has no validation split (Table 8), and the number of clusters k is selected in Table 6 using test-set accuracy; the rank reported for the headline EMTAL-4 result in Table 1 is likewise chosen after inspecting test results. This constitutes test-set tuning. Please use an internal validation split or cross-validation for hyperparameter selection and report the protocol; otherwise the reported gains cannot be separated from selection effects.
  3. [§2.4, Eqs. (10)--(11), Table 4] The Quality Retaining mechanism is the largest single contributor in the ablation (Table 4: 87.39 -> 91.04 with QR alone), but its correctness is unvalidated. The EMA momentum m in Eq. (10) is never reported, and because the bank is updated from iteration 0, it initially contains random logits unless m or a burn-in schedule is chosen to wash them out. In Eq. (11), the weight 1/L_CE,t becomes arbitrarily large as a task converges and is undefined if a mini-batch contains no samples from task t; no batch-construction or loss-scale normalization is specified. Please report m, provide a sensitivity analysis over m and the weighting scheme, and analyze the effect of warm-up/burn-in for Z.
  4. [§2.5, Eq. (12)] The router-fading schedule that guarantees alpha=0 at the end of training is not specified; Section 3.4 only states that the router is faded over 50 epochs. Because the final reparameterized model in Eqs. (13)--(15) is obtained by setting alpha=0, the exact schedule affects both the final accuracy and the efficiency claim. Please give the schedule (e.g., linear or cosine in alpha) and show that the reported performance is insensitive to the choice of schedule.
minor comments (5)
  1. [Abstract, §2.3, Figure 3] There are several typos: 'archiving efficient inference' should be 'achieving', 'fistly' should be 'firstly', and 'fole' in the Figure 3 caption should be 'role'; please copyedit.
  2. [Table 6] The main text lists cluster counts 1, 4, 6, 64, 192, while the table header shows 1, 4, 16, 64, 192; clarify which configuration was actually evaluated.
  3. [Eq. (5)] The indexing E_i^b = E_i^D is inconsistent with the definition of E_i in Eq. (4) and the shape of W in Eq. (3); please clarify the row layout of the concatenated matrix and the bias reshaping.
  4. [§3.2] The hyperparameter tau in Eq. (8) is set to 5 because it is said to exhibit 'stable performance with distinct values', but no sensitivity experiment is shown; either add the experiment or state tau as a default.
  5. [Tables 1 and 2] MOELoRA is reported with 2.82M tunable parameters on Multi-task FGVC but 2.41M on VTAB-1k; please state the LoRA rank and head configuration used for each benchmark.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all central claims are external benchmark measurements; QR's self-distillation is a training mechanism, not a derivation.

full rationale

This is an empirical systems paper: every headline number (Tables 1-3 and 7, Figure 5) is an external benchmark accuracy, and the ablations in Tables 4-6 compare measured accuracies of different configurations. No claimed result is derived from an equation whose inputs already contain the conclusion. The QR mechanism in Eqs. (10)-(11) distills the model toward its own EMA logits; this is a self-referential training objective, but it is not circular reasoning because the paper's claims are about measured downstream accuracy, not about properties entailed by Eq. (10). The MoEfied-LoRA construction, router fading, and weight reparameterization in Eqs. (6)-(15) are structural definitions, and the reported inference-time savings follow from construction, but the accuracy claims are external measurements rather than conclusions loaded into those equations. Citations to prior work, including MoEfication [30,31], LoRA [34], and balanced k-means [37], are independent sources and are not used to assert the paper's empirical superiority. The only self-citations (e.g., [101,102] in the limitations discussion) are non-load-bearing suggestions for future out-of-distribution work. I therefore find no circular step. Separately, the NeurIPS checklist item 7 claims error bars are given in Section 3.5, but Section 3.5 reports Ky Fan norm curves and the accuracy tables are single-run; that is a reporting defect rather than circularity. Hyperparameter choices (cluster count, rank, fading schedule) appear to be selected on the evaluation sets, which is a fitting-to-data concern but does not make any prediction reduce to a fitted input by construction.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claims rest on domain assumptions about weight-similarity clustering producing low-rank experts, EMA logits representing high-quality knowledge, and router fading preserving learned behavior, plus several hand-chosen or unreported hyperparameters. There are no theoretical guarantees; the only evidence is benchmark accuracy.

free parameters (5)
  • Number of clusters k = 16
    Selected by comparing test-set accuracy for k in {1,4,16,64,192} (Table 6, Section 3.4); Multi-task FGVC has no validation split (Table 8), so selection uses the test set.
  • LoRA rank r = 1, 2, 4 (reported separately)
    Rank is chosen per variant EMTAL-1/2/4; authors report all variants rather than one fixed value, and rank affects both accuracy and tunable params.
  • Router temperature tau = 5
    Set in Section 3.2 with the note that it exhibits stable performance across values; effectively a hand-chosen hyperparameter.
  • EMA momentum m = not reported
    Appears in Eq. (10) for the knowledge bank update; its value is never given, though the claim of high-quality logits depends on it.
  • Router-fading alpha schedule = not specified; faded over 50 epochs
    Section 3.4 says the router is diminished over 50 epochs, but the exact schedule for alpha in Eq. (12) is not given.
assumptions (6)
  • domain assumption Similar columns in the FFN up-projection matrix serve similar functions and should be grouped as one expert
    Stated in Section 2.3, borrowed from the key-value memory view (refs 35,36). If false, MoEfied LoRA loses its motivation.
  • domain assumption Clustering weight columns by similarity yields experts with sufficiently low rank that LoRA is effective
    Section 2.3 and the Ky Fan norm figure (Fig 4); the improvement from +LoRA in Table 5 (90.14 vs 89.20) is the only direct evidence.
  • domain assumption The EMA per-class logit bank stores high-quality knowledge whose distillation prevents forgetting
    Section 2.4, Eqs. (10)-(11); not proven, and the bank is seeded with early training logits.
  • domain assumption Fading the router to uniform and discarding it preserves the benefits learned during routing
    Section 2.5 claims the router's knowledge is implicitly absorbed; no mechanism is demonstrated. Table 5 shows accuracy does not drop, but only on this benchmark.
  • standard math Balanced k-means clustering produces a valid partition of the concatenated weight matrix
    Used in Eqs. (3)-(4); relying on the algorithm from ref 37.
  • standard math LoRA updates can be exactly merged into the expert matrices (reparameterization)
    Eqs. (6)-(7) and (13)-(15); this is standard LoRA merging, used to remove router and extra parameters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transforming Vision Transformer: Towards Efficient Multi-Task Asynchronous Learning." pith.science (2026). https://pith.science/paper/NUPMWV3M

@misc{pith2026250106884,
  author       = {Pith},
  title        = {Pith review of: Transforming Vision Transformer: Towards Efficient Multi-Task Asynchronous Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NUPMWV3M}},
  note         = {Machine review of arXiv:2501.06884}
}
read the original abstract

Multi-Task Learning (MTL) for Vision Transformer aims at enhancing the model capability by tackling multiple tasks simultaneously. Most recent works have predominantly focused on designing Mixture-of-Experts (MoE) structures and in tegrating Low-Rank Adaptation (LoRA) to efficiently perform multi-task learning. However, their rigid combination hampers both the optimization of MoE and the ef fectiveness of reparameterization of LoRA, leading to sub-optimal performance and low inference speed. In this work, we propose a novel approach dubbed Efficient Multi-Task Learning (EMTAL) by transforming a pre-trained Vision Transformer into an efficient multi-task learner during training, and reparameterizing the learned structure for efficient inference. Specifically, we firstly develop the MoEfied LoRA structure, which decomposes the pre-trained Transformer into a low-rank MoE structure and employ LoRA to fine-tune the parameters. Subsequently, we take into account the intrinsic asynchronous nature of multi-task learning and devise a learning Quality Retaining (QR) optimization mechanism, by leveraging the historical high-quality class logits to prevent a well-trained task from performance degradation. Finally, we design a router fading strategy to integrate the learned parameters into the original Transformer, archiving efficient inference. Extensive experiments on public benchmarks demonstrate the superiority of our method, compared to the state-of-the-art multi-task learning approaches.

Figures

Figures reproduced from arXiv: 2501.06884 by the authors.

Figure 1
Figure 1. FFN as Mixture of Low-rank Experts. Given an up-projection weight matrix in FFN, a straightforward way of splitting it into MoE is to divide every K channels into separate experts, resulting in highly dissimilar experts and a high-rank MoE, which is inherently unsuitable for integration with LoRA. In contrast, our proposed MoLE approach rearranges the weight matrix into groups of similar channels as experts, creatin… view at source ↗
Figure 2
Figure 2. Summary of representative architectures of multi-task learning. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the proposed EMTAL framework. Given a pre-trained ViT, we firstly [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Comparison of the low-rank properties by using the vanilla MoE and the proposed MoLE, [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Comparison results using various separate training approaches in the context of few-shot [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

119 extracted references · 52 canonical work pages

  1. [1]

    Multi-task learning as multi-objective optimization

    Ozan Sener and Vladlen Koltun. Multi-task learning as multi-objective optimization. In Advances in Neural Information Processing Systems, pages 525–536, 2018

    2018

  2. [2]

    Gradient surgery for multi-task learning

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning. In Advances in Neural Information Processing Systems, pages 5824–5836, 2020

    2020

  3. [3]

    Conflict-averse gradient descent for multi-task learning

    Bo Liu, Xingchao Liu, Xiaojie Jin, Peter Stone, and Qiang Liu. Conflict-averse gradient descent for multi-task learning. In Advances in Neural Information Processing Systems, pages 18878–18890, 2021

    2021

  4. [4]

    M3vit: Mixture-of-experts vision transformer for efficient multi-task learning with model-accelerator co-design

    Zhiwen Fan, Rishov Sarkar, Ziyu Jiang, Tianlong Chen, Kai Zou, Yu Cheng, Cong Hao, Zhangyang Wang, et al. M3vit: Mixture-of-experts vision transformer for efficient multi-task learning with model-accelerator co-design. In Advances in Neural Information Processing Systems, volume 35, pages 28441–28457, 2022

    2022

  5. [5]

    Multi-task dense prediction via mixture of low-rank experts

    Yuqi Yang, Peng-Tao Jiang, Qibin Hou, Hao Zhang, Jinwei Chen, and Bo Li. Multi-task dense prediction via mixture of low-rank experts. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27927–27937, 2024

    2024

  6. [6]

    An erudite fine-grained visual classification model

    Dongliang Chang, Yujun Tong, Ruoyi Du, Timothy Hospedales, Yi-Zhe Song, and Zhanyu Ma. An erudite fine-grained visual classification model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7268–7277, 2023

    2023

  7. [7]

    The kalai- smorodinsky solution for many-objective bayesian optimization

    Mickaël Binois, Victor Picheny, Patrick Taillandier, and Abderrahmane Habbal. The kalai- smorodinsky solution for many-objective bayesian optimization. Journal of Machine Learning Research, 21(150):1–42, 2020

    2020

  8. [8]

    Self-supervised generalisation with meta auxiliary learning

    Shikun Liu, Andrew Davison, and Edward Johns. Self-supervised generalisation with meta auxiliary learning. In Advances in Neural Information Processing Systems, 2019

    2019

Show all 119 references
  1. [9]

    Auxiliary learning by implicit differentiation

    Aviv Navon, Idan Achituve, Haggai Maron, Gal Chechik, and Ethan Fetaya. Auxiliary learning by implicit differentiation. In International Conference on Learning Representations, ICLR 2021, 2021

    2021

  2. [10]

    Adamerging: Adaptive model merging for multi-task learning

    Enneng Yang, Zhenyi Wang, Li Shen, Shiwei Liu, Guibing Guo, Xingwei Wang, and Dacheng Tao. Adamerging: Adaptive model merging for multi-task learning. In International Confer- ence on Learning Representations

  3. [11]

    Towards impartial multi-task learning

    Liyang Liu, Yi Li, Zhanghui Kuang, Jing-Hao Xue, Yimin Chen, Wenming Yang, Qingmin Liao, and Wayne Zhang. Towards impartial multi-task learning. In International Conference on Learning Representations, 2021

    2021

  4. [12]

    Modeling task relationships in multi-task learning with multi-gate mixture-of-experts

    Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H Chi. Modeling task relationships in multi-task learning with multi-gate mixture-of-experts. In Proceedings of ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages 1930–1939, 2018

    1930

  5. [13]

    Mod-squad: Designing mixtures of experts as modular multi-task learners

    Zitian Chen, Yikang Shen, Mingyu Ding, Zhenfang Chen, Hengshuang Zhao, Erik G Learned- Miller, and Chuang Gan. Mod-squad: Designing mixtures of experts as modular multi-task learners. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1...

    2023

  6. [14]

    When moe meets llms: Parameter efficient fine-tuning for multi-task medical ap- plications

    Qidong Liu, Xian Wu, Xiangyu Zhao, Yuanshao Zhu, Derong Xu, Feng Tian, and Yefeng Zheng. When moe meets llms: Parameter efficient fine-tuning for multi-task medical ap- plications. In Proceedings of the International ACM SIGIR Conference on Research and Development in Informat...

    2024

  7. [15]

    Adaptive mixtures of local experts

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts. Neural Computation, 3(1):79–87, 1991. 11

    1991

  8. [16]

    Learning piecewise control strategies in a modular neural network architecture

    Robert A Jacobs and Michael I Jordan. Learning piecewise control strategies in a modular neural network architecture. IEEE Transactions on Systems, Man, and Cybernetics, 23(2):337– 345, 1993

    1993

  9. [17]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. arXiv preprint arXiv:1701.06538, 2017

    2017 arXiv

  10. [18]

    Belongie, Bharath Hariha- ran, and Ser-Nam Lim

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge J. Belongie, Bharath Hariha- ran, and Ser-Nam Lim. Visual prompt tuning. In Proceedings of the European Conference on Computer Vision, pages 709–727, 2022

    2022

  11. [19]

    Scaling & shifting your features: A new baseline for efficient model tuning

    Dongze Lian, Daquan Zhou, Jiashi Feng, and Xinchao Wang. Scaling & shifting your features: A new baseline for efficient model tuning. In Advances in Neural Information Processing Systems, pages 109–123, 2022

    2022

  12. [20]

    Efficient adaptation of large vision transformer via adapter re-composing

    Wei Dong, Dawei Yan, Zhijun Lin, and Peng Wang. Efficient adaptation of large vision transformer via adapter re-composing. In Advances in Neural Information Processing Systems, 2023

    2023

  13. [21]

    Compressed video prompt tuning.Advances in Neural Information Processing Systems, 36:31895–31907, 2023

    Bing Li, Jiaxin Chen, Xiuguo Bao, and Di Huang. Compressed video prompt tuning.Advances in Neural Information Processing Systems, 36:31895–31907, 2023

    2023

  14. [22]

    Gradnorm: Gra- dient normalization for adaptive loss balancing in deep multitask networks

    Zhao Chen, Vijay Badrinarayanan, Chen-Yu Lee, and Andrew Rabinovich. Gradnorm: Gra- dient normalization for adaptive loss balancing in deep multitask networks. In International Conference on Machine Learning, pages 794–803, 2018

    2018

  15. [23]

    Gradient vaccine: Investigating and improving multi-task optimization in massively multilingual models

    Zirui Wang, Yulia Tsvetkov, Orhan Firat, and Yuan Cao. Gradient vaccine: Investigating and improving multi-task optimization in massively multilingual models. arXiv preprint arXiv:2010.05874, 2020

    2010 arXiv

  16. [24]

    Multi-task learning as a bargaining game

    Aviv Navon, Aviv Shamsian, Idan Achituve, Haggai Maron, Kenji Kawaguchi, Gal Chechik, and Ethan Fetaya. Multi-task learning as a bargaining game. In International Conference on Machine Learning, volume 162, pages 16428–16446, 2022

    2022

  17. [25]

    Independent component alignment for multi-task learning

    Dmitry Senushkin, Nikolay Patakin, Arseny Kuznetsov, and Anton Konushin. Independent component alignment for multi-task learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20083–20093, 2023

    2023

  18. [26]

    Multinet++: Multi-stream feature aggregation and geometric loss strategy for multi-task learning

    Sumanth Chennupati, Ganesh Sistu, Senthil Yogamani, and Samir A Rawashdeh. Multinet++: Multi-stream feature aggregation and geometric loss strategy for multi-task learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019

    2019

  19. [27]

    Baijiong Lin, Feiyang Ye, Yu Zhang, and Ivor W. Tsang. Reasonable effectiveness of random weighting: A litmus test for multi-task learning. Transactions on Machine Learning Research, 2022, 2022

    2022

  20. [28]

    End-to-end multi-task learning with attention

    Shikun Liu, Edward Johns, and Andrew J Davison. End-to-end multi-task learning with attention. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1871–1880, 2019

    2019

  21. [29]

    Achievement-based training progress balancing for multi-task learning

    Hayoung Yun and Hanjoo Cho. Achievement-based training progress balancing for multi-task learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 16935–16944, 2023

    2023

  22. [30]

    Moefication: Transformer feed-forward layers are mixtures of experts

    Zhengyan Zhang, Yankai Lin, Zhiyuan Liu, Peng Li, Maosong Sun, and Jie Zhou. Moefication: Transformer feed-forward layers are mixtures of experts. In Proceedings of Annual Meeting of the Association for Computational Linguistics, pages 877–890, 2022

    2022

  23. [31]

    Exploiting transformer activation sparsity with dynamic inference

    Mikołaj Piórczy ´nski, Filip Szatkowski, Klaudia Bałazy, and Bartosz Wójcik. Exploiting transformer activation sparsity with dynamic inference. arXiv preprint arXiv:2310.04361, 2023. 12

    2023 arXiv

  24. [32]

    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...

    2021

  25. [33]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages 5998–6008, 2017

    2017

  26. [34]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022

    2022

  27. [35]

    Transformer feed-forward layers are key-value memories

    Mor Geva, Roei Schuster, Jonathan Berant, and Omer Levy. Transformer feed-forward layers are key-value memories. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5484–5495, 2021

    2021

  28. [36]

    Knowledge neurons in pretrained transformers

    Damai Dai, Li Dong, Yaru Hao, Zhifang Sui, Baobao Chang, and Furu Wei. Knowledge neurons in pretrained transformers. In Proceedings of Annual Meeting of the Association for Computational Linguistics, pages 8493–8502, 2022

    2022

  29. [37]

    Balanced k-means for clustering

    Mikko I Malinen and Pasi Fränti. Balanced k-means for clustering. In Structural, Syntactic, and Statistical Pattern Recognition: Joint IAPR International Workshop, pages 32–41, 2014

    2014

  30. [38]

    Distilling the knowledge in a neural network

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

    2015 arXiv

  31. [39]

    The caltech-ucsd birds-200-2011 dataset

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Perona, and Serge Belongie. The caltech-ucsd birds-200-2011 dataset. 2011

    2011

  32. [40]

    Fine- grained car detection for visual census estimation

    Timnit Gebru, Jonathan Krause, Yilun Wang, Duyun Chen, Jia Deng, and Li Fei-Fei. Fine- grained car detection for visual census estimation. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4502–4508, 2017

    2017

  33. [41]

    Fine- grained visual classification of aircraft

    Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew Blaschko, and Andrea Vedaldi. Fine- grained visual classification of aircraft. arXiv preprint arXiv:1306.5151, 2013

    2013 arXiv

  34. [42]

    Automated flower classification over a large number of classes

    Maria-Elena Nilsback and Andrew Zisserman. Automated flower classification over a large number of classes. In Indian Conference on Computer Vision, Graphics & Image Processing, pages 722–729, 2008

    2008

  35. [43]

    A large-scale study of representation learning with the visual task adaptation benchmark

    Xiaohua Zhai, Joan Puigcerver, Alexander Kolesnikov, Pierre Ruyssen, Carlos Riquelme, Mario Lucic, Josip Djolonga, Andre Susano Pinto, Maxim Neumann, Alexey Dosovitskiy, et al. A large-scale study of representation learning with the visual task adaptation benchmark. arXiv prep...

    1910 arXiv

  36. [44]

    Indoor segmentation and support inference from rgbd images

    Pushmeet Kohli Nathan Silberman, Derek Hoiem and Rob Fergus. Indoor segmentation and support inference from rgbd images. In Proceedings of the European Conference on Computer Vision, 2012

    2012

  37. [45]

    Taskprompter: Spatial-channel multi-task prompting for dense scene understanding

    Hanrong Ye and Dan Xu. Taskprompter: Spatial-channel multi-task prompting for dense scene understanding. In International Conference on Learning Representations, 2023

    2023

  38. [46]

    Learning to prompt for vision-language models

    Kaiyang Zhou, Jingkang Yang, Chen Change Loy, and Ziwei Liu. Learning to prompt for vision-language models. International Journal of Computer Vision, 130(9):2337–2348, 2022

    2022

  39. [47]

    Neural prompt search

    Yuanhan Zhang, Kaiyang Zhou, and Ziwei Liu. Neural prompt search. arXiv preprint arXiv:2206.04673, 2022

    2022 arXiv

  40. [48]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In International Conference on Learning Representations, 2019. 13

    2019

  41. [49]

    Dualprompt: Complementary prompting for rehearsal-free continual learning

    Zifeng Wang, Zizhao Zhang, Sayna Ebrahimi, Ruoxi Sun, Han Zhang, Chen-Yu Lee, Xiaoqi Ren, Guolong Su, Vincent Perot, Jennifer Dy, et al. Dualprompt: Complementary prompting for rehearsal-free continual learning. In Proceedings of the European Conference on Computer Vision, pag...

    2022

  42. [50]

    Parameter-efficient transfer learning for NLP

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin de Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. Parameter-efficient transfer learning for NLP. In International Conference on Machine Learning, pages 2790–2799, 2019

    2019

  43. [51]

    Sensitivity-aware visual parameter-efficient fine-tuning

    Haoyu He, Jianfei Cai, Jing Zhang, Dacheng Tao, and Bohan Zhuang. Sensitivity-aware visual parameter-efficient fine-tuning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 11825–11835, 2023

    2023

  44. [52]

    Xuan Vinh Doan and Stephen A. Vavasis. Low-rank matrix recovery with ky fan 2-k-norm. J. Glob. Optim., 82(4):727–751, 2022

    2022

  45. [53]

    Multitask learning

    Rich Caruana. Multitask learning. Machine Learning, 28:41–75, 1997

    1997

  46. [54]

    Multi-task learning with deep neural networks: A survey

    Michael Crawshaw. Multi-task learning with deep neural networks: A survey. arXiv preprint arXiv:2009.09796, 2020

    2009 arXiv

  47. [55]

    An overview of multi-task learning in deep neural networks

    Sebastian Ruder. An overview of multi-task learning in deep neural networks. arXiv preprint arXiv:1706.05098, 2017

    2017 arXiv

  48. [56]

    Multi-task learning for dense prediction tasks: A survey

    Simon Vandenhende, Stamatios Georgoulis, Wouter Van Gansbeke, Marc Proesmans, Dengxin Dai, and Luc Van Gool. Multi-task learning for dense prediction tasks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(7):3614–3633, 2021

    2021

  49. [57]

    Nddr-cnn: Layerwise feature fusing in multi-task cnns by neural discriminative dimensionality reduction

    Yuan Gao, Jiayi Ma, Mingbo Zhao, Wei Liu, and Alan L Yuille. Nddr-cnn: Layerwise feature fusing in multi-task cnns by neural discriminative dimensionality reduction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3205–3214, 2019

    2019

  50. [58]

    Cross-stitch networks for multi-task learning

    Ishan Misra, Abhinav Shrivastava, Abhinav Gupta, and Martial Hebert. Cross-stitch networks for multi-task learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3994–4003, 2016

    2016

  51. [59]

    Latent multi- task architecture learning

    Sebastian Ruder, Joachim Bingel, Isabelle Augenstein, and Anders Søgaard. Latent multi- task architecture learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 4822–4829, 2019

    2019

  52. [60]

    Automated search for resource-efficient branched multi-task networks

    David Bruggemann, Menelaos Kanakis, Stamatios Georgoulis, and Luc Van Gool. Automated search for resource-efficient branched multi-task networks. arXiv preprint arXiv:2008.10292, 2020

    2008 arXiv

  53. [61]

    Learning to branch for multi-task learning

    Pengsheng Guo, Chen-Yu Lee, and Daniel Ulbricht. Learning to branch for multi-task learning. In International Conference on Machine Learning, pages 3854–3863, 2020

    2020

  54. [62]

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics

    Alex Kendall, Yarin Gal, and Roberto Cipolla. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 7482–7491, 2018

    2018

  55. [63]

    Exploring relational context for multi-task dense prediction

    David Brüggemann, Menelaos Kanakis, Anton Obukhov, Stamatios Georgoulis, and Luc Van Gool. Exploring relational context for multi-task dense prediction. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15869–15878, 2021

    2021

  56. [64]

    Mti-net: Multi-scale task interaction networks for multi-task learning

    Simon Vandenhende, Stamatios Georgoulis, and Luc Van Gool. Mti-net: Multi-scale task interaction networks for multi-task learning. In Proceedings of the European Conference on Computer Vision, pages 527–543, 2020

    2020

  57. [65]

    Pattern-structure diffusion for multi-task learning

    Ling Zhou, Zhen Cui, Chunyan Xu, Zhenyu Zhang, Chaoqun Wang, Tong Zhang, and Jian Yang. Pattern-structure diffusion for multi-task learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4514–4523, 2020. 14

    2020

  58. [66]

    Pattern-affinitive propagation across depth, surface normal and semantic segmentation

    Zhenyu Zhang, Zhen Cui, Chunyan Xu, Yan Yan, Nicu Sebe, and Jian Yang. Pattern-affinitive propagation across depth, surface normal and semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4106–4115, 2019

    2019

  59. [67]

    Joint task- recursive learning for semantic segmentation and depth estimation

    Zhenyu Zhang, Zhen Cui, Chunyan Xu, Zequn Jie, Xiang Li, and Jian Yang. Joint task- recursive learning for semantic segmentation and depth estimation. In Proceedings of the European Conference on Computer Vision, pages 235–251, 2018

    2018

  60. [68]

    Robust learning through cross-task consistency

    Amir R Zamir, Alexander Sax, Nikhil Cheerla, Rohan Suri, Zhangjie Cao, Jitendra Malik, and Leonidas J Guibas. Robust learning through cross-task consistency. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11197–11206, 2020

    2020

  61. [69]

    Learning to prompt for continual learning

    Zifeng Wang, Zizhao Zhang, Chen-Yu Lee, Han Zhang, Ruoxi Sun, Xiaoqi Ren, Guolong Su, Vincent Perot, Jennifer Dy, and Tomas Pfister. Learning to prompt for continual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 139–149, 2022

    2022

  62. [70]

    Just pick a sign: Optimizing deep multitask models with gradient sign dropout

    Zhao Chen, Jiquan Ngiam, Yanping Huang, Thang Luong, Henrik Kretzschmar, Yuning Chai, and Dragomir Anguelov. Just pick a sign: Optimizing deep multitask models with gradient sign dropout. In Advances in Neural Information Processing Systems , volume 33, pages 2039–2050, 2020

    2020

  63. [71]

    Dynamic task prioritization for multitask learning

    Michelle Guo, Albert Haque, De-An Huang, Serena Yeung, and Li Fei-Fei. Dynamic task prioritization for multitask learning. In Proceedings of the European conference on computer vision, pages 270–287, 2018

    2018

  64. [72]

    A modulation module for multi-task learning with applications in image retrieval

    Xiangyun Zhao, Haoxiang Li, Xiaohui Shen, Xiaodan Liang, and Ying Wu. A modulation module for multi-task learning with applications in image retrieval. In Proceedings of the European Conference on Computer Vision, pages 401–416, 2018

    2018

  65. [73]

    Contrastive multi-task dense prediction

    Siwei Yang, Hanrong Ye, and Dan Xu. Contrastive multi-task dense prediction. InProceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 3190–3197, 2023

    2023

  66. [74]

    Learning multiple dense prediction tasks from partially annotated data

    Wei-Hong Li, Xialei Liu, and Hakan Bilen. Learning multiple dense prediction tasks from partially annotated data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18879–18889, 2022

    2022

  67. [75]

    Asynchronous convergence in multi-task learning via knowledge distillation from converged tasks

    Weiyi Lu, Sunny Rajagopalan, Priyanka Nigam, Jaspreet Singh, Xiaodi Sun, Yi Xu, Belinda Zeng, and Trishul Chilimbi. Asynchronous convergence in multi-task learning via knowledge distillation from converged tasks. In NAACL, pages 149–159, 2022

    2022

  68. [76]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity

    William Fedus, Barret Zoph, and Noam Shazeer. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. Journal of Machine Learning Research, 23(120):1–39, 2022

    2022

  69. [77]

    Gshard: Scaling giant models with conditional computation and automatic sharding

    Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. Gshard: Scaling giant models with conditional computation and automatic sharding. arXiv preprint arXiv:2006.16668, 2020

    2006 arXiv

  70. [78]

    Taming sparsely activated transformer with stochastic experts

    Simiao Zuo, Xiaodong Liu, Jian Jiao, Young Jin Kim, Hany Hassan, Ruofei Zhang, Tuo Zhao, and Jianfeng Gao. Taming sparsely activated transformer with stochastic experts. arXiv preprint arXiv:2110.04260, 2021

    2021 arXiv

  71. [79]

    Taskexpert: Dynamically assembling multi-task representations with memorial mixture-of-experts

    Hanrong Ye and Dan Xu. Taskexpert: Dynamically assembling multi-task representations with memorial mixture-of-experts. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 21828–21837, 2023

    2023

  72. [80]

    Fact: Factor-tuning for lightweight adaptation on vision transformer

    Shibo Jie and Zhi-Hong Deng. Fact: Factor-tuning for lightweight adaptation on vision transformer. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 37, pages 1060–1068, 2023

    2023

  73. [81]

    Polyhistor: Parameter- efficient multi-task adaptation for dense vision tasks

    Yen-Cheng Liu, Chih-Yao Ma, Junjiao Tian, Zijian He, and Zsolt Kira. Polyhistor: Parameter- efficient multi-task adaptation for dense vision tasks. volume 35, pages 36889–36901, 2022. 15

    2022

  74. [82]

    Vl-adapter: Parameter-efficient transfer learning for vision-and-language tasks

    Yi-Lin Sung, Jaemin Cho, and Mohit Bansal. Vl-adapter: Parameter-efficient transfer learning for vision-and-language tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5227–5237, 2022

    2022

  75. [83]

    Intrinsic dimensionality explains the effectiveness of language model fine-tuning

    Armen Aghajanyan, Luke Zettlemoyer, and Sonal Gupta. Intrinsic dimensionality explains the effectiveness of language model fine-tuning. arXiv preprint arXiv:2012.13255, 2020

    2012 arXiv

  76. [84]

    Low-rank compression of neural nets: Learning the rank of each layer

    Yerlan Idelbayev and Miguel A Carreira-Perpinán. Low-rank compression of neural nets: Learning the rank of each layer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8049–8059, 2020

    2020

  77. [85]

    Multi-task learning with low rank attribute embedding for person re-identification

    Chi Su, Fan Yang, Shiliang Zhang, Qi Tian, Larry S Davis, and Wen Gao. Multi-task learning with low rank attribute embedding for person re-identification. In Proceedings of the IEEE International Conference on Computer Vision, pages 3739–3747, 2015

    2015

  78. [86]

    Generalized low rank models

    Madeleine Udell, Corinne Horn, Reza Zadeh, Stephen Boyd, et al. Generalized low rank models. Foundations and Trends® in Machine Learning, 9(1):1–118, 2016

    2016

  79. [87]

    Trace norm regularised deep multi-task learning

    Yongxin Yang and Timothy M Hospedales. Trace norm regularised deep multi-task learning. arXiv preprint arXiv:1606.04038, 2016

    2016 arXiv

  80. [88]

    Lo- rahub: Efficient cross-task generalization via dynamic lora composition

    Chengsong Huang, Qian Liu, Bill Yuchen Lin, Tianyu Pang, Chao Du, and Min Lin. Lo- rahub: Efficient cross-task generalization via dynamic lora composition. arXiv preprint arXiv:2307.13269, 2023

    2023 arXiv

  81. [89]

    Mixture of cluster-conditional lora experts for vision-language instruction tuning

    Yunhao Gou, Zhili Liu, Kai Chen, Lanqing Hong, Hang Xu, Aoxue Li, Dit-Yan Yeung, James T Kwok, and Yu Zhang. Mixture of cluster-conditional lora experts for vision-language instruction tuning. arXiv preprint arXiv:2312.12379, 2023

    2023 arXiv

  82. [90]

    Loramoe: Revolutionizing mixture of experts for maintaining world knowledge in language model alignment

    Shihan Dou, Enyu Zhou, Yan Liu, Songyang Gao, Jun Zhao, Wei Shen, Yuhao Zhou, Zhi- heng Xi, Xiao Wang, Xiaoran Fan, et al. Loramoe: Revolutionizing mixture of experts for maintaining world knowledge in language model alignment. arXiv preprint arXiv:2312.09979, 2023

    2023 arXiv

  83. [91]

    Octavius: Mitigating task interference in mllms via lora-moe

    Zeren Chen, Ziqin Wang, Zhen Wang, Huayang Liu, Zhenfei Yin, Si Liu, Lu Sheng, Wanli Ouyang, and Jing Shao. Octavius: Mitigating task interference in mllms via lora-moe. In International Conference on Learning Representations, 2024

    2024

  84. [92]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. Transactions on Machine Learning Research, 2024, 2024

    2024

  85. [93]

    Bitfit: Simple parameter-efficient fine-tuning for transformer-based masked language-models

    Elad Ben Zaken, Yoav Goldberg, and Shauli Ravfogel. Bitfit: Simple parameter-efficient fine-tuning for transformer-based masked language-models. In Proceedings of Annual Meeting of the Association for Computational Linguistics, pages 1–9, 2022

    2022

  86. [94]

    One-for-all: Generalized lora for parameter-efficient fine-tuning

    Arnav Chavan, Zhuang Liu, Deepak Gupta, Eric Xing, and Zhiqiang Shen. One-for-all: Generalized lora for parameter-efficient fine-tuning. arXiv preprint arXiv:2306.07967, 2023

    2023 arXiv

  87. [95]

    Controlling text-to-image diffusion by orthogonal finetuning

    Zeju Qiu, Weiyang Liu, Haiwen Feng, Yuxuan Xue, Yao Feng, Zhen Liu, Dan Zhang, Adrian Weller, and Bernhard Schölkopf. Controlling text-to-image diffusion by orthogonal finetuning. In Advances in Neural Information Processing Systems, pages 79320–79362, 2023

    2023

  88. [96]

    Parameter-efficient orthogonal finetuning via butterfly factorization

    Weiyang Liu, Zeju Qiu, Yao Feng, Yuliang Xiu, Yuxuan Xue, Longhui Yu, Haiwen Feng, Zhen Liu, Juyeon Heo, Songyou Peng, et al. Parameter-efficient orthogonal finetuning via butterfly factorization. In International Conference on Learning Representations, 2024

    2024

  89. [97]

    Rota- tion equivariant cnns for digital pathology

    Bastiaan S Veeling, Jasper Linmans, Jim Winkens, Taco Cohen, and Max Welling. Rota- tion equivariant cnns for digital pathology. In International Conference on Medical Image Computing and Computer-Assisted Intervention, 2018

    2018

  90. [98]

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification

    Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019. 16

    2019

  91. [99]

    Remote sensing image scene classification: Benchmark and state of the art

    Gong Cheng, Junwei Han, and Xiaoqiang Lu. Remote sensing image scene classification: Benchmark and state of the art. Proceedings of the IEEE, 2017

    2017

  92. [100]

    Kaggle diabetic retinopathy detection, July 2015

    Kaggle and EyePacs. Kaggle diabetic retinopathy detection, July 2015

    2015

  93. [101]

    Target-relevant knowledge preservation for multi-source domain adaptive object detection

    Jiaxi Wu, Jiaxin Chen, Mengzhe He, Yiru Wang, Bo Li, Bingqi Ma, Weihao Gan, Wei Wu, Yali Wang, and Di Huang. Target-relevant knowledge preservation for multi-source domain adaptive object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recog...

    2022

  94. [102]

    Domain adaptive object detection for uav-based images by robust representation learning and multiple pseudo-label aggregation

    Wu Ke, Jiaxin Chen, and Miao Wang. Domain adaptive object detection for uav-based images by robust representation learning and multiple pseudo-label aggregation. In Proceedings of the 1st International Workshop on Efficient Multimedia Computing under Limited, pages 59–67, 2024

    2024

  95. [103]

    Wilds: A benchmark of in-the-wild distribution shifts

    Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, et al. Wilds: A benchmark of in-the-wild distribution shifts. In International Conference on Machine Learning, p...

    2021

  96. [104]

    LoRAMoE [90] splits the LoRA experts into two groups and explicitly learns distinct capabilities for each group

    designs a router network that is conditioned on the clustering information extracted from each input sample. LoRAMoE [90] splits the LoRA experts into two groups and explicitly learns distinct capabilities for each group. While these mixture-of-LoRA methods densely combine mul...

    2011

  97. [105]

    Guidelines: • The answer NA means that the abstract and introduction do not include the claims made in the paper

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer:[Yes] Justification: We describe it in Section 1. Guidelines: • The answer NA means that the abstract and introduction do not include the cl...

  98. [106]

    Limitations

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: We describe it in Section E. Guidelines: • The answer NA means that the paper has no limitation while the answer No means that the paper has limitati...

  99. [107]

    Guidelines: • The answer NA means that the paper does not include theoretical results

    Theory Assumptions and Proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] 22 Justification: No theoretical result. Guidelines: • The answer NA means that the paper does not include t...

  100. [108]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental Result Reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main ex- perimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and da...

  101. [109]

    Guidelines: • The answer NA means that paper does not include experiments requiring code

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instruc- tions to faithfully reproduce the main experimental results, as described in supplemental material? 23 Answer: [No] Justification: We describe details of pu...

  102. [110]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental Setting/Details Question: Does the paper specify all the training and test details (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: We describe it in Section D. ...

  103. [111]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiment Statistical Significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [Yes] Justification: We describe it in Section 3.5. Guidelines: • The a...

  104. [112]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiments Compute Resources Question: For each experiment, does the paper provide sufficient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: We describe it in Secti...

  105. [113]

    Guidelines: • The answer NA means that the authors have not reviewed the NeurIPS Code of Ethics

    Code Of Ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: The research conducted in the paper conform, in every respect, with the NeurIPS...

  106. [114]

    Guidelines: • The answer NA means that there is no societal impact of the work performed

    Broader Impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: We discuss it in Section E. Guidelines: • The answer NA means that there is no societal impact of the wor...

  107. [115]

    Guidelines: • The answer NA means that the paper poses no such risks

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: The paper poses...

  108. [116]

    Guidelines: • The answer NA means that the paper does not use existing assets

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: The creators or ...

  109. [117]

    Guidelines: • The answer NA means that the paper does not release new assets

    New Assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [NA] Justification: The paper does not release new assets. Guidelines: • The answer NA means that the paper does not release new assets. ...

  110. [118]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Crowdsourcing and Research with Human Subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  111. [119]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional Review Board (IRB) Approvals or Equivalent for Research with Human Subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

Pith tools

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