Pith. sign in

REVIEW 3 major objections 6 minor 42 references

EVA: Mixture-of-Experts Semantic Variant Alignment for Compositional Zero-Shot Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read EVA claims top results in compositional zero-shot learning with MoE adapters and variant alignment

desk verdict A plausible and cleanly written method with solid internal ablations, but the central SOTA claim is undermined by comparing against baselines run on a smaller backbone. read the letter →

arxiv 2506.20986 v1 pith:YHOST4Y2 submitted 2025-06-26 cs.CV

Xiao Zhang , Yongqiang Ma , Haodong Jing , Nanning Zheng This is my paper · ORCID
classification cs.CV
keywords compositionalzero-shotlearningmixture-of-expertssemanticvariantalignmentvision-languagemodelsCLIPLoRAadapteropen-worldrecognitionfine-grainedattribute-object
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 tries to establish that compositional zero-shot learning—recognizing unseen state-object pairs such as "blue grass" from learned state and object concepts—is improved by replacing the uniform composition-to-prototype mapping used in prior work with two mechanisms: token-aware mixture-of-experts adapters and semantic variant alignment. On MIT-States, UT-Zappos, and C-GQA, the proposed EVA framework reports AUC gains of +1.5, +5.7, and +4.4 points in the closed-world setting, and +1.2, +2.9, and +1.2 points in the open-world setting. If these results hold, they would show that expert specialization and instance-dependent primitive matching, built on frozen CLIP encoders, are effective ingredients for compositional generalization.

What carries the argument

The central mechanism is the MoE adapter combined with the variant-selection alignment. The adapter, placed at every encoder layer, uses a router to assign each token to one shared expert and K routed experts; the experts are LoRA low-rank projections, keeping the parameter cost small while letting different experts specialize in state-related or object-related semantic content. The semantic variant alignment then operates globally to locally: from the text view, primitive probabilities are computed by taking the maximum over composition probabilities sharing that primitive; from the image view, the CLS tokens from all experts in the final image layer are treated as alternative visual descriptions, and the variant with the highest inter-model (variant-to-primitive) plus intra-model (variant-to-composition) affinity score is chosen as the primitive visual feature. This replaces the all-to-one image-primitive alignment that the paper argues disrupts the topological structure of cross-modal fine-grained associations.

What would settle it

Rerun the cited baselines (DFSP, Troika, CDS-CZSL, RAPR, and others) with the same frozen CLIP ViT-L/14 backbone and the same 20-epoch training protocol used for EVA; if their AUC and harmonic-mean scores rise to match EVA's, the central claim of superiority is falsified.

Watch

Extended reading notes

Core claim

The central claim is that primitive representations in CZSL should be modeled as clusters of semantic variants, not single centroids, and that the tokens carrying those variants should be routed to specialized experts. EVA inserts LoRA-based MoE adapters in parallel with the feed-forward network of every layer of the frozen CLIP image and text encoders, with a shared expert capturing common knowledge and K=2 routed experts chosen per token by a router. For alignment, text-to-image matching scores a state or object by the maximum composition score over all compositions sharing that primitive, while image-to-text matching treats the final-layer CLS tokens from the different experts as feature variants and selects the variant with highest combined affinity to the corresponding primitive text feature and to the composition feature. The paper reports state-of-the-art results on all three benchmarks in both closed- and open-world settings, with the largest margins on UT-Zappos (AUC 50.2 vs. 44.5 for the best prior method) and C-GQA (AUC 18.8 vs. 14.4).

Load-bearing premise

The comparison assumes that the published baseline numbers, produced with the smaller CLIP ViT-B/16 encoder, are directly comparable to EVA's results using the larger frozen CLIP ViT-L/14, and that the reported margin reflects the proposed components rather than the backbone.

Editorial extensions

If this is right

  • If the reported numbers are correct, token-level MoE adapters are a parameter-efficient way to adapt frozen CLIP encoders for CZSL without adding suffix modules or unfreezing the backbone.
  • The variant-based alignment implies that primitive matching should depend on the instance, not on a single fixed prototype, which is a different inductive bias from most existing CZSL alignments.
  • The larger gains on UT-Zappos and C-GQA suggest the benefit grows with fine-grained label spaces and larger composition sets.
  • The open-world AUC gains of +1.2 to +2.9 indicate that the alignment transfers to the full Cartesian-product space, not just the curated test split.
  • The ablations attribute most of the improvement to domain-expert adaption, with semantic variant alignment adding a further gain on top, supporting the claim that both components are needed.

Reading between the lines

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

  • Because the paper uses the larger CLIP ViT-L/14 backbone while the cited baselines report results with ViT-B/16, a matched-backbone rerun of the baselines is needed to separate method gains from backbone gains; the stated margins may shrink once baselines are rerun under the same backbone.
  • The implicit assumption that compositions sharing a primitive form a cluster centered on the primitive text feature is testable on other attribute-object datasets and on retrieval tasks, where the selected variant could be evaluated directly against ground-truth attributes.
  • The expert-load analysis suggests the router learns a division of labor between state and object domains; extending the same adapter design to other vision-language problems that require attribute-object disentanglement may be a direct payoff.
  • Scaling the number of experts or using patch-level rather than CLS-token variants could push the variant-selection mechanism further, since its capacity is bounded by the number and diversity of expert outputs.
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

3 major / 6 minor

Summary. The paper proposes EVA, a compositional zero-shot learning method built on frozen CLIP encoders. It inserts LoRA-based Mixture-of-Experts adapters into the image and text encoders (domain-expert adaption) and adds two alignment regularizers: text-to-image alignment aggregates composition probabilities into state and object scores, and image-to-text alignment selects among the encoder's expert feature variants using inter- and intra-model affinity. The training objective combines composition, state, and object cross-entropy losses, and inference adds the state and object scores to the composition logits. The paper reports state-of-the-art results on MIT-States, UT-Zappos, and C-GQA in both closed- and open-world settings, with ablations on the two main components.

Significance. The proposed architecture is conceptually clean, and the internal ablation (Table 4) is consistent: both domain-expert adaption and semantic variant alignment improve over a same-backbone prompt-only baseline, and the gains co-occur. If the claims were established under a matched backbone, the token-level MoE adapter and variant-selection alignment would be a useful contribution to CZSL. However, the headline comparison to previous methods is not currently valid because the cited baselines were evaluated with a weaker visual backbone than EVA. The paper's claim to state-of-the-art performance is therefore not established by the present evidence; the method's internal gains are real but smaller than the reported SOTA margins. The paper also does not report error bars or release code, which limits verification of the empirical claims.

major comments (3)
  1. [Section 4.1; Tables 1 and 3] EVA is built on frozen CLIP ViT-L/14 (Section 4.1, Implementation Details), while the published numbers quoted for DFSP, CDS-CZSL, Troika, PLID, RAPR, and the other baselines were obtained with the CLIP ViT-B/16 backbone. The paper neither reruns any baseline with ViT-L/14 nor reports EVA with ViT-B/16. This makes the claimed closed-world AUC gains (+1.5%, +5.7%, +4.4%) and open-world AUC gains (+1.2%, +2.9%, +1.2%) not attributable to the proposed method. Table 4 supports this concern: the prompt-only BASELINE with ViT-L/14 already reaches an AUC of 20.2 on MIT-States, which is essentially at the level of published ViT-B/16 DFSP (20.6–20.8) and close to CDS-CZSL (22.4). Please provide same-backbone comparisons, either by rerunning the baselines under ViT-L/14 or by training EVA under ViT-B/16, before the SOTA claim can be evaluated.
  2. [Section 3.4, Eqs. (13)–(14)] The image-to-text alignment selects a feature variant by arg max over the affinity scores A_S and A_O, but the paper does not state how gradients are passed through this discrete selection for the losses L_v^s and L_v^o in Eqs. (17)–(18). Since the selected variant is used as f_s and f_o in the subsequent probability computations, the training dynamics depend crucially on this implementation (e.g., straight-through estimation, a softmax relaxation, or detached routing). Please specify and justify the differentiable implementation. If a hard arg max is used with no gradient to non-selected experts, the router and expert outputs in the image encoder receive no learning signal from this loss, which should be discussed explicitly.
  3. [Tables 1 and 3; Section 4.2] The paper reports single-run numbers with no standard deviations or number of seeds. Many of the decisive margins are small in the harder settings, such as open-world C-GQA (AUC 5.6 vs. 4.4 for RAPR) and MIT-States closed-world (AUC 24.0 vs. 22.5 for CDS-CZSL). Given the backbone mismatch and the absence of variance estimates, the claim of 'significantly outperforms other state-of-the-art CZSL methods' is not supported. Please report mean and standard deviation over at least three seeds for EVA and, if feasible, for the baselines rerun under the same backbone.
minor comments (6)
  1. [Eqs. (13)–(14)] The notation is imprecise: f_s and f_o are written as the result of an arg max over affinity scores, but in Eqs. (15)–(16) they are used as feature vectors. Write f_s = v_{i^*_s} with i^*_s = arg max_i a^s_i, and similarly for f_o.
  2. [Section 4.2, Open-World paragraph] The text refers to 'RAPR [21]' when comparing on UT-Zappos; RAPR is reference [13], while [21] is CDS-CZSL. Please correct the citation.
  3. [Section 4.1 vs. Table 2] The prose states that C-GQA has 453 states and 870 objects, while Table 2 lists 413 states and 674 objects. Please reconcile this discrepancy.
  4. [Section 4.3] The phrase 'to access the effect of key components' should be 'to assess the effect'; this appears to be a typo.
  5. [Section 4.1; Eqs. (7)–(8)] The trainable coefficients tau_s and tau_o in Eqs. (7)–(8) are not given initial values or constraints in Section 4.1. Please specify them, since they directly scale the state and object probabilities used in the loss.
  6. [Table 2] The numbers in Table 2 run together in the rendered text (e.g., '281751262' and '278362'); add clear column separators so the dataset split statistics are readable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: EVA's losses are consistency regularizers over the model's own scores, not fitted-to-target predictions; the backbone mismatch is an experimental fairness concern, not a circularity.

full rationale

EVA's derivation chain does not reduce to its inputs. The text-to-image alignment in Eqs. (7)-(8) defines p_s and p_o as max-aggregations of the model's own composition probabilities p_c, and Eq. (20) combines these aggregates with p_c at inference; this is a consistency re-ranking of the same model scores, not a quantity fitted to the test set or a prediction that is definitionally equal to its training target. The image-to-text alignment (Eqs. (11)-(18)) selects among the model's own MoE variants using affinity scores and then trains cross-entropy losses against ground-truth state/object labels, which is standard auxiliary supervision rather than circular self-confirmation. The method is trained only on seen compositions, with hyperparameters (λ1, λ2, α, β, r, K) set by validation or ablation, and no parameter is fitted to the test set. The paper does not rely on a self-citation chain: the cited prior work (DFSP, Troika, CDS-CZSL, RAPR, etc.) is external and used for comparison, not as an unverified load-bearing premise. The ablation in Table 4 shows each component improves over a same-backbone baseline, giving independent internal support. The reviewer concern about CLIP ViT-L/14 versus the baselines' ViT-B/16 is an experimental fairness/comparability issue, not a circularity, and therefore does not raise the circularity score.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a set of hyperparameters and assumptions about CLIP feature quality and MoE specialization. The most notable missing piece is a description of the differentiable variant selection, which is required for training. No fundamentally new entities are introduced; the 'experts' are an established MoE mechanism.

free parameters (8)
  • alpha = 0.5
    Balancing coefficient in the intra-model affinity score (Eq. 12), chosen by hand.
  • lambda1 = 0.5
    Weight for the text-to-image state/object losses (Eq. 19), tuned on validation.
  • lambda2 = 0.1
    Weight for image-to-text primitive losses (Eq. 19), tuned on validation.
  • beta = 0.5
    Inference combination weight between composition and primitive scores (Eq. 20).
  • tau_s = not reported
    Trainable temperature scaling for state probabilities (Eq. 7); final value not reported.
  • tau_o = not reported
    Trainable temperature scaling for object probabilities (Eq. 8); final value not reported.
  • number_of_activated_experts_K = 2
    Chosen by ablation on C-GQA (Table 5b).
  • expert_hidden_dim_r = 64
    Chosen by ablation on C-GQA (Table 5a).
assumptions (4)
  • domain assumption Frozen CLIP ViT-L/14 provides semantically meaningful image and text representations for compositional reasoning.
    The entire method builds on CLIP features (Section 3.3) and does not verify that its representations are valid for CZSL beyond the empirical results.
  • domain assumption Compositions can be seen as a cluster of semantic variants centered on their primitive features, so that max over composition scores approximates state/object scores.
    This motivates Eqs. (7)-(8) and the variant selection in Section 3.4.
  • domain assumption MoE routers can learn to allocate tokens to experts that specialize in state-like versus object-like knowledge.
    The paper relies on this for 'domain-expert adaption' to improve primitive modeling (Section 3.3).
  • ad hoc to paper The arg-max selection in Eq. (13)-(14) is implemented in a differentiable way that provides useful gradients.
    The paper does not describe how gradients flow through the hard selection; this is essential for training but unstated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EVA: Mixture-of-Experts Semantic Variant Alignment for Compositional Zero-Shot Learning." pith.science (2026). https://pith.science/paper/YHOST4Y2

@misc{pith2026250620986,
  author       = {Pith},
  title        = {Pith review of: EVA: Mixture-of-Experts Semantic Variant Alignment for Compositional Zero-Shot Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YHOST4Y2}},
  note         = {Machine review of arXiv:2506.20986}
}
read the original abstract

Compositional Zero-Shot Learning (CZSL) investigates compositional generalization capacity to recognize unknown state-object pairs based on learned primitive concepts. Existing CZSL methods typically derive primitives features through a simple composition-prototype mapping, which is suboptimal for a set of individuals that can be divided into distinct semantic subsets. Moreover, the all-to-one cross-modal primitives matching neglects compositional divergence within identical states or objects, limiting fine-grained image-composition alignment. In this study, we propose EVA, a Mixture-of-Experts Semantic Variant Alignment framework for CZSL. Specifically, we introduce domain-expert adaption, leveraging multiple experts to achieve token-aware learning and model high-quality primitive representations. To enable accurate compositional generalization, we further present semantic variant alignment to select semantically relevant representation for image-primitives matching. Our method significantly outperforms other state-of-the-art CZSL methods on three popular benchmarks in both closed- and open-world settings, demonstrating the efficacy of the proposed insight.

Figures

Figures reproduced from arXiv: 2506.20986 by the authors.

Figure 1
Figure 1. Previous primitive alignment forces the composition vi [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a). The framework of EVA consists of domain-expert adaption for token-aware representation learning and semantic variant alignment for fine-grained image-primitives matching. (b). Domain expert adaption leverages MoE adapter to dynamically process seman￾tically relevant tokens with in-domain knowledge. (c). Semantic variant alignment introduces text-to-image and image-to-text alignment to select most relevant featu… view at source ↗
Figure 3
Figure 3. Qualitative Results. We present top-3 predictions of randomly selected images in terms of success (Left) and failure cases (Right). [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Left: Visualization of image features learned by BASELINE and our method. Right: The token load of various experts in state and object domains. sults are achieved when all components are utilized, demon￾strating the effectiveness of the proposed method. 4.4. Qualitativ…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 23 canonical work pages

  1. [1]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    arXiv

  2. [2]

    Learning to generalize to new compositions in image understanding

    Yuval Atzmon, Jonathan Berant, Vahid Kezami, Amir Globerson, and Gal Chechik. Learning to generalize to new compositions in image understanding.arXiv preprint arXiv:1608.07639, 2016. 1

    work page Pith review arXiv 2016

  3. [3]

    A causal view of compositional zero-shot recognition.Ad- vances in Neural Information Processing Systems, 33:1462– 1473, 2020

    Yuval Atzmon, Felix Kreuk, Uri Shalit, and Gal Chechik. A causal view of compositional zero-shot recognition.Ad- vances in Neural Information Processing Systems, 33:1462– 1473, 2020. 2

    work page 2020

  4. [4]

    Prompting language-informed distribution for compositional zero-shot learning

    Wentao Bao, Lichang Chen, Heng Huang, and Yu Kong. Prompting language-informed distribution for compositional zero-shot learning. InEuropean Conference on Computer Vision, pages 107–123. Springer, 2024. 1, 2, 5, 6

    work page 2024

  5. [5]

    Lan- guage models are few-shot learners.Advances in neural in- formation processing systems, 33:1877–1901, 2020

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Lan- guage models are few-shot learners.Advances in neural in- formation processing systems, 33:1877–1901, 2020. 2

    1901

  6. [6]

    Bert: Pre-training of deep bidirectional trans- formers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional trans- formers for language understanding. InNorth American Chapter of the Association for Computational Linguistics, 2019

    work page 2019

  7. [7]

    Unified language model pre-training for natural language un- derstanding and generation.Advances in neural information processing systems, 32, 2019

    Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language un- derstanding and generation.Advances in neural information processing systems, 32, 2019. 2

    work page 2019

  8. [8]

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

    J. Edward Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models.ArXiv, abs/2106.09685, 2021. 3, 6

    arXiv 2021

Show all 42 references
  1. [9]

    Troika: Multi-path cross-modal trac- tion for compositional zero-shot learning

    Siteng Huang, Biao Gong, Yutong Feng, Min Zhang, Yiliang Lv, and Donglin Wang. Troika: Multi-path cross-modal trac- tion for compositional zero-shot learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 24005–24014, 2024. 1...

    2024

  2. [10]

    Dis- covering states and transformations in image collections

    Phillip Isola, Joseph J Lim, and Edward H Adelson. Dis- covering states and transformations in image collections. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1383–1391, 2015. 2, 5

    2015

  3. [11]

    Jacobs, Michael I

    Robert A. Jacobs, Michael I. Jordan, Steven J. Nowlan, and Geoffrey E. Hinton. Adaptive mixtures of local experts.Neu- ral Computation, 3:79–87, 1991. 2

    1991

  4. [12]

    Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Deven- dra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guil- laume Lample, L ´elio Renard Lavaud, Lucile Saulnier, Mari...

    2024 arXiv

  5. [13]

    Retrieval-augmented primitive representations for composi- tional zero-shot learning

    Chenchen Jing, Yukun Li, Hao Chen, and Chunhua Shen. Retrieval-augmented primitive representations for composi- tional zero-shot learning. InProceedings of the AAAI Con- ference on Artificial Intelligence, pages 2652–2660, 2024. 1, 2, 5, 6

    2024

  6. [14]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    arXiv

  7. [15]

    PhD thesis, Massachusetts Institute of Technology,

    Brenden M Lake.Towards more human-like concept learn- ing in machines: Compositionality, causality, and learning- to-learn. PhD thesis, Massachusetts Institute of Technology,

  8. [16]

    Building machines that learn and think like people.Behavioral and brain sciences, 40:e253,

    Brenden M Lake, Tomer D Ullman, Joshua B Tenenbaum, and Samuel J Gershman. Building machines that learn and think like people.Behavioral and brain sciences, 40:e253,

  9. [17]

    Align before fuse: Vision and language representation learn- ing with momentum distillation.Advances in neural infor- mation processing systems, 34:9694–9705, 2021

    Junnan Li, Ramprasaath Selvaraju, Akhilesh Gotmare, Shafiq Joty, Caiming Xiong, and Steven Chu Hong Hoi. Align before fuse: Vision and language representation learn- ing with momentum distillation.Advances in neural infor- mation processing systems, 34:9694–9705, 2021. 2

    2021

  10. [18]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. InInterna- tional conference on machine learning, pages 12888–12900. PMLR, 2022. 2

    2022

  11. [19]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. InIn- ternational conference on machine learning, pages 19730– 19742. PMLR, 2023. 2

    2023

  12. [20]

    Siamese contrastive embedding network for composi- tional zero-shot learning

    Xiangyu Li, Xu Yang, Kun Wei, Cheng Deng, and Muli Yang. Siamese contrastive embedding network for composi- tional zero-shot learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9326–9335, 2022. 1

    2022

  13. [21]

    Context-based and diversity-driven specificity in compositional zero-shot learning

    Yun Li, Zhe Liu, Hang Chen, and Lina Yao. Context-based and diversity-driven specificity in compositional zero-shot learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17037– 17046, 2024. 1, 5, 6

    2024

  14. [22]

    Symme- try and group in attribute-object compositions

    Yong-Lu Li, Yue Xu, Xiaohan Mao, and Cewu Lu. Symme- try and group in attribute-object compositions. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11316–11325, 2020. 1, 2

    2020

  15. [23]

    Deepseek-v3 technical report

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437, 2024. 2

    2024 arXiv

  16. [24]

    Visual instruction tuning.Advances in neural information processing systems, 36, 2024

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Advances in neural information processing systems, 36, 2024. 2

    2024

  17. [25]

    De- composed soft prompt guided fusion enhancing for compo- sitional zero-shot learning

    Xiaocheng Lu, Song Guo, Ziming Liu, and Jingcai Guo. De- composed soft prompt guided fusion enhancing for compo- sitional zero-shot learning. InProceedings of the IEEE/CVF 9 Conference on Computer Vision and Pattern Recognition, pages 23560–23569, 2023. 1, 2, 3, 4, 5, 6

    2023

  18. [26]

    Open world compositional zero- shot learning

    Massimiliano Mancini, Muhammad Ferjad Naeem, Yongqin Xian, and Zeynep Akata. Open world compositional zero- shot learning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5222– 5230, 2021. 2, 3, 4

    2021

  19. [27]

    Learning graph embeddings for open world compositional zero-shot learning.IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2022

    Massimiliano Mancini, Muhammad Ferjad Naeem, Yongqin Xian, and Zeynep Akata. Learning graph embeddings for open world compositional zero-shot learning.IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2022. 2, 3

    2022

  20. [28]

    Efficient estimation of word representa- tions in vector space.arXiv preprint arXiv:1301.3781, 3781,

    Tomas Mikolov. Efficient estimation of word representa- tions in vector space.arXiv preprint arXiv:1301.3781, 3781,

    arXiv

  21. [29]

    From red wine to red tomato: Composition with context

    Ishan Misra, Abhinav Gupta, and Martial Hebert. From red wine to red tomato: Composition with context. InProceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1792–1801, 2017. 1, 2, 3, 4, 5

    2017

  22. [30]

    Learning graph embeddings for compositional zero-shot learning

    Muhammad Ferjad Naeem, Yongqin Xian, Federico Tombari, and Zeynep Akata. Learning graph embeddings for compositional zero-shot learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 953–962, 2021. 1, 2, 3, 5, 6

    2021

  23. [31]

    Learning to compose soft prompts for compositional zero-shot learning

    Nihal V Nayak, Peilin Yu, and Stephen H Bach. Learning to compose soft prompts for compositional zero-shot learning. arXiv preprint arXiv:2204.03574, 2022. 2, 5, 6

    2022 arXiv

  24. [32]

    Pytorch: An im- perative style, high-performance deep learning library.Ad- vances in neural information processing systems, 32, 2019

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An im- perative style, high-performance deep learning library.Ad- vances in neural information processing systems, ...

    2019

  25. [33]

    Task-driven modular networks for zero-shot compositional learning

    Senthil Purushwalkam, Maximilian Nickel, Abhinav Gupta, and Marc’Aurelio Ranzato. Task-driven modular networks for zero-shot compositional learning. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 3593–3602, 2019. 2

    2019

  26. [34]

    Language models are unsu- pervised multitask learners.OpenAI blog, 1(8):9, 2019

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsu- pervised multitask learners.OpenAI blog, 1(8):9, 2019. 2

    2019

  27. [35]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. InInternational Conference on Machine Learning,...

    2021

  28. [36]

    Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc V

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

    arXiv

  29. [37]

    Gemini 1.5: Unlocking multimodal understanding across millions of tokens of con- text.arXiv preprint arXiv:2403.05530, 2024

    Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, Zhufeng Pan, Shibo Wang, et al. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of con- text.arXiv preprint arXiv:2403.05530, 2024. 2

    2024 arXiv

  30. [38]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 2

    2017

  31. [39]

    Prompting large pre-trained vision-language models for compositional concept learning.arXiv preprint arXiv:2211.05077, 2022

    Guangyue Xu, Parisa Kordjamshidi, and Joyce Chai. Prompting large pre-trained vision-language models for compositional concept learning.arXiv preprint arXiv:2211.05077, 2022. 1, 2, 5, 6

    2022 arXiv

  32. [40]

    Gipcol: Graph-injected soft prompting for compositional zero-shot learning

    Guangyue Xu, Joyce Chai, and Parisa Kordjamshidi. Gipcol: Graph-injected soft prompting for compositional zero-shot learning. InProceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision, pages 5774–5783,

  33. [41]

    Fine-grained visual compar- isons with local learning

    Aron Yu and Kristen Grauman. Fine-grained visual compar- isons with local learning. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 192–199, 2014. 2, 5

    2014

  34. [42]

    Learning to prompt for vision-language models.In- ternational Journal of Computer Vision, 130(9):2337–2348,

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

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.