Pith. sign in

REVIEW 3 major objections 6 minor 62 references

MASCOT: Model-Aware Submodular Coverage for Composite-Attribute Text-to-Image Retrieval

T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Coverage-based re-ranking keeps early-rank recall that manifold repulsion loses on diversity-decrease tasks.

desk verdict A genuinely honest and useful paper on diversity-decrease re-ranking whose headline MS-DPP comparison is undercut by unreported baseline tuning, but the coverage-vs-repulsion mechanism is worth taking seriously. read the letter →

arxiv 2608.12532 v1 pith:EHUTOAAN submitted 2026-08-12 cs.MM cs.CVcs.IR

Aaryan Sharma , Vishak Prasad C , Virendra Singh , Ganesh Ramakrishnan This is my paper · ORCID
classification cs.MMcs.CVcs.IR
keywords text-to-imageretrievalresultdiversificationsubmodularoptimizationvision-languagemodelsdeterminantalpointprocessessoftbinningmetadata-awarere-rankingearly-rankrecall
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

This paper claims that the standard manifold-repulsion approach to result diversification is structurally fragile in the direction most systems actually want: deliberately narrowing retrieval along discrete metadata such as hour of day or geographic region. When diversity must be decreased, MS-DPP's continuous repulsion discards semantically relevant images and early-rank recall collapses—on the composite geo+hour task R@10 falls from 0.9737 to 0.4931 and R@1 to 0.2346. The proposed alternative, MASCOT, treats diversity as bounded coverage over soft metadata bins, weighting each bin by the query relevance it contains, so the penalty for concentrating results decays as the target range saturates. Averaged across the three decrease tasks MASCOT retains R@10 = 0.8858 against MS-DPP's 0.6763, and on the composite task it holds R@10 = 0.9410 and R@1 = 0.7202. The paper is explicit that this is not uniform dominance: its own simpler ablations score higher harmonic means on all three decrease tasks, and the advantage is specific to early-rank recall under composite constraints.

What carries the argument

The carrying object is the soft Information Unit (IU) with its saturation-bounded marginal gain. Discrete metadata are mapped into a probability distribution over coarse bins—24 hourly bins for time, a $g\times g$ degree-space grid for geography—via a Gaussian kernel $p(u,i)=\exp\left(-\mathrm{dist}(a_i,\mathrm{center}(u))^2/2\sigma^2\right)$, so an image at 12:59 and one at 1:01 partially share coverage instead of being split by a boundary. The objective's directional term is the weighted coverage $d\lambda\sum_u \Omega(u,q)\left(1-\prod_i(1-p(u,i))\right)$, where $\Omega(u,q)=\max_j p(u,j)\hat{R}(j,q)$ makes bin importance query-driven, and the greedy marginal gain $\Delta_{\mathrm{cov}}(i)=\sum_u \Omega(u)(1-P_{\mathrm{covered},u})p(u,i)$ is the mechanism: as $P_{\mathrm{covered},u}\to 1$, the factor $(1-P_{\mathrm{covered},u})$ drives the gain—and, under $d=-1$, the penalty—to zero, so selection reverts to the locally normalized relevance $\hat{R}$. In increase mode the objective is monotone submodular and greedy selection carries the classical $(1-1/e)$ approximation guarantee; in decrease mode it is a difference-of-submodular objective and the same greedy loop is treated as a validated heuristic.

What would settle it

Re-run the PP_geo_hour diversity-decrease experiment on the full 25,151-image PixelProse pool (or a fresh download of the current release) at the paper's reported settings, and check whether MASCOT's R@10 around 0.94 versus MS-DPP's around 0.49 persists; if the gap closes at full scale, the reported advantage is an artifact of the 996-image surviving subset. A secondary check uses the paper's own sensitivity sweep: R@10 on that task falls from 0.9410 at $\lambda=0.1$ to 0.0477 at $\lambda=0.2$, so testing whether the margin survives a hyperparameter selection that leaves room away from the collapse boundary would settle whether the method is robustly better or tuned to the cliff edge.

Watch

Extended reading notes

Core claim

The paper's central discovery is that diversity-decrease on discrete metadata is a saturation problem, not a compression problem, and that coverage-based selection solves it where volume-minimizing repulsion cannot. Determinantal selection penalizes an image for being distinct from the already-selected set with no floor, so under a decrease constraint it keeps punishing relevant images even after the target metadata range is densely covered; MASCOT's coverage penalty, by contrast, is proportional to the residual capacity of the bins an image occupies, so once a bin saturates the penalty vanishes and the objective reverts to normalized semantic relevance. Concretely, MASCOT projects each image into a Gaussian soft-binning space of Information Units, weighs each unit by $\Omega(u,q)$, the maximum relevance-weighted occupancy it receives from the candidate set, and greedily maximizes $f(S)=(1-\lambda)\sum_{i\in S}\hat{R}(i,q) + d\lambda\sum_{u\in U}\Omega(u,q)\left(1-\prod_{i\in S}(1-p(u,i))\right)$, with $d=-1$ converting coverage into a bounded penalty. The result is a recall curve that loses a bounded amount of relevance at rank 1 and recovers quickly with rank—R@10 = 0.9410 on the composite PP_geo_hour decrease task versus MS-DPP's 0.4931—rather than a curve that stays depressed across the whole window. The paper also characterizes the boundary of its own claim: the top-ranked result is still displaced (the first selection matches the base retriever on 84.4%, 76.0% and 65.8% of queries across the three decrease tasks), and on aggregate diversity-relevance scores the Uniform Binning ablation is better on all three decrease tasks.

Load-bearing premise

The headline numbers come from 996 PixelProse images that survived EXIF filtering and URL downloads, and the paper assumes this pool is representative enough that relative method rankings transfer to the original 25,151-image benchmark—if URL attrition or metadata filtering shifted task difficulty, the measured R@10 gap could be specific to the surviving subset rather than intrinsic to the methods.

Editorial extensions

If this is right

  • On diversity-decrease tasks over discrete metadata, a coverage re-ranker preserves far more early-rank relevance than manifold repulsion, and the margin grows when two attributes must be suppressed at once: R@10 0.9410 vs 0.4931 on PP_geo_hour.
  • Because the penalty is bounded by residual bin capacity, the recall lost at rank 1 is bounded and recovered quickly with rank (R@1 0.7202 recovering to 0.9410 by rank 10), whereas MS-DPP's recall stays depressed across the whole curve.
  • Query-driven bin weighting is not an aggregate improvement: on all three decrease tasks the Uniform Binning ablation attains higher harmonic means, so $\Omega$'s contribution is the shape of the recall curve—mid-rank recovery under composite constraints at the cost of early-rank accuracy.
  • The approach only pays off with capture-accurate metadata and a candidate pool large enough for repulsion to overreach; on small pools (VG_hour) or IP-geolocated coordinates (I1M_geo) MS-DPP leads or ties.
  • Mixed-direction control (increase along one attribute, decrease along another) works in the partitioned bin space and beats MS-DPP on both mixed configurations (R@10 0.960 vs 0.916 for geo-up/time-down at $K=10$).

Reading between the lines

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

  • The saturation property is a general design principle, not a specific algorithm: any decrease-mode objective whose penalty is capped by coverage of already-selected items should resist early-rank collapse, so DPP-style methods with a capped or regularized repulsion term are a natural testable extension.
  • The reported operating points sit one step from abrupt collapse (R@10 falls from 0.9410 at $\lambda=0.1$ to 0.0477 at $\lambda=0.2$ on PP_geo_hour decrease), so real deployments would need the paper's one-step-probe rule automated into hyperparameter selection; otherwise the gains are a knife-edge.
  • The 996-image evaluation pool is the untested hinge: a full-scale re-run on the original 25,151-image PixelProse pool could shrink or widen the measured gap, and would be the cleanest way to decide whether the R@10 advantage is intrinsic or a property of the surviving subset.
  • Because the objective is attribute-agnostic (the paper shows cluster-derived Information Units work), the same coverage formulation could control learned or semantic attributes—style, scene type, sentiment—where repulsion methods lack a discrete axis; whether early-rank preservation survives there is an open question.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes MASCOT, a submodular-coverage re-ranker for text-to-image retrieval over composite metadata attributes. MASCOT projects discrete metadata into soft Information Units, weights each unit by query-driven relevance, and greedily maximizes a weighted sum of normalized relevance and a directional coverage term. The authors report that on PixelProse diversity-decrease tasks, MASCOT preserves early-rank recall much better than the manifold-repulsion baseline MS-DPP, especially on the composite PP_geo_hour task, while candidly noting that its own Uniform Binning ablation achieves higher harmonic means on the same decrease tasks and that the advantage is confined to mid-rank recovery under composite constraints. Extensive appendices provide sensitivity sweeps, dataset-specific results, a runtime comparison, and an explicit statement of operating regimes and limitations.

Significance. If the central empirical claim holds, MASCOT offers a useful alternative mechanism for controllable diversification: a coverage-based objective whose decrease-mode penalty saturates as bins fill, rather than an unbounded repulsive penalty. The paper is unusually candid: it reports ablation failures, acknowledges that its simpler ablation beats it on aggregate harmonic means, gives a full sensitivity analysis with abrupt-collapse boundaries, and releases code. The main weakness is that the headline comparison against MS-DPP depends on an operating point that is not shown to be fair or fully swept, and the matched-recall analysis in the appendix partly undercuts the headline claim. The contribution is nevertheless plausible and the empirical apparatus is reproducible in principle.

major comments (3)
  1. [§7.1, Table 1; Appendix D.6, Table 10; Table 5] The headline comparison is not yet supported because the MS-DPP operating point is not demonstrated to be fair. Table 10 reports that MS-DPP's best R@10 on PP_geo_hour decrease is 0.4931, yet at zero diversification intensity MS-DPP should reduce to the base BLIP-2 ranking with R@10=0.9737, exactly as the paper's own lambda=0 row in Table 6 does for MASCOT. The dash for MS-DPP in Table 10 therefore implies that the baseline sweep never approached the no-diversification limit, or that its intensity/weight parameters were constrained away from zero. Table 5 lists validation-selected hyperparameters only for MASCOT and its ablations, not for MS-DPP or the other baselines, so the reader cannot verify the sweep. Please report the full lambda/w_j sweep for MS-DPP and all variants, include the lambda=0 sanity point, and compare methods at matched diversity metric or matched recall rather than at arbitrary selected operating points.
  2. [Appendix D.6, Table 10; §7.1, Table 1] The advantage over manifold baselines is not robust to operating-point matching. Table 10 shows that at R@10 >= 0.90 on PP_geo_hour decrease, MS-DPP+TN+TVMS achieves DM=0.2066 while MASCOT achieves DM=0.1881, and the paper itself concedes that MASCOT is not the strongest compressor at matched recall. The abstract and Section 7.1 nonetheless frame the result as a recall collapse of MS-DPP relative to MASCOT using only the vanilla variant. Please add a matched-DM comparison across all MS-DPP variants, or explicitly narrow the claim to the specific vanilla configuration and state the matched-recall trade-off in the main text.
  3. [§6.1, Dataset and Tasks] The PixelProse evaluation uses 996 retained images out of the 25,151 reported in the original MS-DPP benchmark, with the difference attributed to EXIF filtering and URL attrition. The statement that relative comparisons remain valid justifies internal comparison among methods on the same pool, but it does not establish that the absolute recall levels or the magnitude of the MS-DPP failure transfer to the original 25,151-image benchmark. If URL attrition or EXIF filtering systematically selects certain cameras, regions, or time distributions, the central gap could be specific to the filtered pool. Please report the metadata distribution of the retained versus discarded images, and if possible run a robustness check on a larger or stratified sample, or on the original split.
minor comments (6)
  1. [Algorithm 1, line 14] The symbol rendered as "˝" in the marginal coverage gain formula appears to be a typographical corruption of the summation sign; please replace it with a proper \sum_{u \in U}.
  2. [§6.1 and Appendix G.5] The setup says 1,000 dense captions are randomly sampled, but Appendix G.5 refers to 797 test queries; please clarify whether the 1,000 includes validation captions, or whether some sampled captions are dropped.
  3. [Appendix D.1, Table 5 caption] The caption states that Table 5 lists validation-selected hyperparameters for every method triple reported in the main tables, but the table contains only MASCOT and its ablations; please reword the caption and either add the baseline hyperparameters or point to where they are reported.
  4. [Appendix D.6, Table 10] The dash for MS-DPP is unverifiable without the sweep grid; please state the exact lambda values and per-attribute weights w_j used so that the claim that no operating point meets the R@10 floors can be checked.
  5. [Section 4.5 and Appendix G.5] The contribution list says the paper "characterizes when" the first selection preserves or displaces the top-ranked result, but Appendix G.5 provides only empirical agreement rates and a classification of displacements; please either add a formal condition in terms of lambda, Omega, and p, or soften the wording to "empirically characterizes."
  6. [§7.1] The statement that MS-DPP+TN+TVMS "struggles to meaningfully improve" DM (0.2066 versus 0.1656) is somewhat subjective, since the difference is a roughly 25% relative increase; please either add a statistical comparison or use more neutral wording.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: MASCOT's objective is self-contained, the Omega-Rhat correlation is disclosed as a design property, and the headline claims are empirical comparisons against external baselines.

full rationale

MASCOT's objective (Eq. 5) combines normalized BLIP-2 relevance Rhat (Eq. 3) with query-driven bin importance Omega (Eq. 4), where Omega is explicitly defined as max_j p(u,j)*Rhat(j,q). This means the coverage term and the relevance term are correlated by construction, as the paper openly states in Section 4.5 and Appendix G.5: 'The two terms are correlated by construction: Omega is built from Rhat... This is the mechanism behind the top-1 displacement quantified in the appendix.' That is a disclosed design property and a mechanism hypothesis, not an unacknowledged reduction of a prediction to its fitted input. The central comparative claims, MASCOT R@10 = 0.9410 versus MS-DPP 0.4931 on PP_geo_hour decrease and the averaged 88.58% versus 67.63%, are measured outcomes on a held-out test split (Section 6, Table 1) against external baselines, with hyperparameters tuned on validation (Appendix D, Table 5). The paper also candidly reports that its own Uniform Binning ablation achieves higher harmonic means on all three decrease tasks and that the advantage disappears on VG_hour and I1M_geo (Appendix F). The Omega-Rhat coupling and the early-rank behavior are consequences of the objective's definition rather than a fitted rename, and the paper's own limitation statements, including the abrupt collapse one grid step past the selected operating point (Appendix D), undercut any suggestion that the result is forced by construction. The remaining concerns, such as the 996-image PixelProse subset and whether MS-DPP was swept to a near-no-diversification operating point (Table 10: 'MS-DPP's best R@10 on this task is 0.4931'), are benchmarking-validity questions rather than circularity: they do not show that MASCOT's derivation reduces to its own inputs. No self-citation chain, uniqueness theorem, or ansatz-smuggling citation is load-bearing, so the paper is self-contained with respect to the circularity criterion.

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

No new entities are postulated. Information Units are adopted from Xu et al. [50], and soft-binning uses a standard Gaussian kernel. All invented content is methodological (the query-driven weighting and direction-dependent coverage), not ontological.

free parameters (4)
  • lambda (diversification intensity) = PP_geo dec 0.3, PP_hour dec 0.4, PP_geo_hour dec 0.1; inc 0.8-0.9
    Validation grid search selects lambda per task and direction; the reported decrease operating points sit at the largest lambda before recall collapse (Appendix D.2).
  • sigma_geo (soft-binning bandwidth, degrees) = 10.0 (PP_geo dec), 15.0 (PP_geo_hour dec), 1.0-15.0 range
    Validation grid search; sweep shows recall collapses from 0.8105 to 0.0013 when moving from 10 to 15 degrees on PP_geo dec.
  • sigma_time (soft-binning bandwidth, hours) = 0.5 (PP_hour dec), 3.0 (PP_geo_hour dec), 0.5-1.5 increase
    Validation grid search; PP_hour dec falls from 0.9059 to 0.0351 when increasing from 0.5 to 1.0 hours.
  • geographic grid resolution g = 20 (PP_geo and PP_geo_hour)
    Validation-selected; R@10 falls from 0.8105 to 0.0025 at g=25 on PP_geo dec, so the selected value is one step before collapse.
assumptions (5)
  • standard math The probabilistic coverage function cov(S) is monotone submodular.
    Used in Section 5 to claim the (1-1/e) greedy guarantee for increase mode; property established in Xu et al. [50].
  • standard math The decrease-mode objective f- = (1-lambda)R - lambda*cov is a difference of submodular functions, so greedy selection is treated as a heuristic without an approximation guarantee.
    Section 5 explicitly states the (1-1/e) bound does not apply for decrease mode.
  • domain assumption Planar degree-space Euclidean distance is a valid proxy for geographic soft-binning; antimeridian wrap and latitude-dependent longitude contraction are negligible.
    Section 4.1 asserts no query straddles +/-180 degrees and high-latitude fractions are small (0.4%, 0.6%, 4.6%) across splits; a geodesic kernel is acknowledged as the principled choice for globally uniform data.
  • domain assumption Relative comparisons on the 996 retained PixelProse images remain valid for benchmarking re-rankers.
    Section 6.1 acknowledges the gap from 25,151 images and asserts the same filtered set is used for all methods; this assumes URL attrition is not correlated with task difficulty.
  • domain assumption The Diversity Metric (DM) of Sogi et al. is a valid evaluation metric, and the appearance channel is near-constant across methods so DM differences reflect metadata concentration.
    Appendix G.6/G.3 reports img_vendi within a 0.0072 band and uses this to interpret DM.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MASCOT: Model-Aware Submodular Coverage for Composite-Attribute Text-to-Image Retrieval." pith.science (2026). https://pith.science/paper/EHUTOAAN

@misc{pith2026260812532,
  author       = {Pith},
  title        = {Pith review of: MASCOT: Model-Aware Submodular Coverage for Composite-Attribute Text-to-Image Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EHUTOAAN}},
  note         = {Machine review of arXiv:2608.12532}
}
read the original abstract

Vision-Language Models (VLMs) are highly effective in retrieving semantically relevant images. However, in practice, relevance alone is often insufficient. Systems must also achieve Result Diversification (RD) across composite attributes such as geography and time, a task for which precise control remains challenging. Current re-ranking methods, such as Multi-Source Determinantal Point Processes (MS-DPP), address this using manifold-based repulsion over similarity representations. Although this strategy is effective for broad exploration, it exposes a key limitation in manifold-based models: when subjected to diversity-decrease tasks on discrete metadata, they suffer substantial degradation in early-rank recall. To bridge this gap, we introduce MASCOT (Model-Aware Submodular Coverage for Composite-Attribute Text-to-Image Retrieval). Instead of relying on manifold repulsion, MASCOT formulates multi-attribute diversity as a resource allocation problem, projecting attributes into a soft-binning space weighted by query-driven importance. Averaged across the three PixelProse diversity-decrease tasks, MASCOT preserves an early-rank recall (R@10) of 88.58%, while MS-DPP retains 67.63%. The margin widens under composite constraints: on PP_geo_hour, where temporal and geographic diversity must be suppressed simultaneously, MS-DPP's recall collapses from 0.9737 to 0.4931 and its top-ranked result degrades to R@1 = 0.23, while MASCOT holds R@10 = 0.9410 and R@1 = 0.7202 at a diversity metric above the unconstrained baseline. We do not claim uniform superiority: on aggregate diversity-relevance scores our own simpler ablations attain higher harmonic means on all three decrease tasks, and MASCOT's advantage is specific to recall beyond rank 1 under composite constraints.

Figures

Figures reproduced from arXiv: 2608.12532 by the authors.

Figure 1
Figure 1. Recall@K curves (𝐾 ∈ [1, 20]) across all datasets and tasks. Top Row (Increase): Continuous manifold repulsion remains stable when expanding the retrieval space, with most models closely tracking the BLIP-2 baseline. Bottom Row (Decrease): When forced to tightly bound attributes, manifold models (MS-DPP variants) and older algorithms (k-DPP, Clustering) suffer severe early-rank degradation on the PixelProse tasks, d… view at source ↗
Figure 2
Figure 2. Recall@K curves (𝐾 ∈ {1, 3, 5, 10, 20}) on the VG_hour and I1M_geo benchmarks. Left column (Increase): All methods track closely, consistent with the PixelProse findings; spatial repulsion and coverage-based models alike preserve semantic recall during diversity expansion. Right column (Decrease): The severity of the manifold vulnerability depends on dataset characteristics, and on VG_hour it does not appear at all.… view at source ↗
Figure 3
Figure 3. Qualitative comparison of retrieval under composite-attribute refinement on [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of retrieval under composite-attribute refinement on [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 24 canonical work pages

  1. [1]

    Approximate Inference in Continuous Determinantal Point Processes

    Raja Hafiz Affandi, Emily B. Fox, and Ben Taskar. 2013. Approximate Inference in Continuous Determinantal Point Processes. arXiv:1311.2971 [stat.ML] https: //arxiv.org/abs/1311.2971

    work page Pith review arXiv 2013

  2. [2]

    Rakesh Agrawal, Sreenivas Gollapudi, Alan Halverson, and Samuel Ieong. 2009. Diversifying search results. InProceedings of the Second ACM International Con- ference on Web Search and Data Mining(Barcelona, Spain)(WSDM ’09). Associa- tion for Computing Machinery, New York, NY, USA, 5–14. doi:10.1145/1498759. 1498766

    work page doi:10.1145/1498759 2009

  3. [3]

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katie Millican, Malcolm Reynolds, Roman Ring, Eliza Rutherford, Serkan Cabi, Tengda Han, Zhitao Gong, Sina Samangooei, Marianne Monteiro, Jacob Menick, Sebastian Borgeaud, Andrew Brock, Aida Nematzadeh, Sahand Sharifzadeh, Mikolaj Binkowski,...

    arXiv 2022

  4. [4]

    Andrew An Bian, Baharan Mirzasoleiman, Joachim Buhmann, and Andreas Krause. 2017. Guaranteed non-convex optimization: Submodular maximization over continuous domains. InArtificial Intelligence and Statistics. PMLR, 111–120

    work page 2017

  5. [5]

    C.M. Bishop. 2006.Pattern recognition and machine learning. Vol. 4. Springer New York. http://scholar.google.com/scholar.bib?q=info:jYxggZ6Ag1YJ:scholar. google.com/&output=citation&hl=en&as_sdt=0,5&as_vis=1&ct=citation&cd=0

    work page 2006

  6. [6]

    Bogdan Boteanu, Ionuţ Mironică, and Bogdan Ionescu. 2017. Pseudo-relevance feedback diversification of social image retrieval results.Multimedia Tools and Applications76, 9 (2017), 11889–11916

    work page 2017

  7. [7]

    Gabriele Capannini, Franco Maria Nardini, Raffaele Perego, and Fabrizio Silvestri

    work page

  8. [8]

    Jaime Carbonell and Jade Goldstein. 1998. The use of MMR, diversity-based reranking for reordering documents and producing summaries. InProceedings of the 21st Annual International ACM SIGIR Conference on Research and Develop- ment in Information Retrieval(Melbourne, Australia)(SIGIR ’98). Association for Computing Machinery, New York, NY, USA, 335–336. d...

    arXiv 1998

Show all 62 references
  1. [9]

    Laming Chen, Guoxin Zhang, and Eric Zhou. 2018. Fast greedy map inference for determinantal point process to improve recommendation diversity.Advances in neural information processing systems31 (2018)

    2018

  2. [10]

    Ruoyu Chen, Hua Zhang, Siyuan Liang, Jingzhi Li, and Xiaochun Cao. 2024. Less is more: Fewer interpretable region via submodular subset selection. In International Conference on Learning Representations, Vol. 2024. 18280–18302

    2024

  3. [11]

    Yen-Chun Chen, Linjie Li, Licheng Yu, Ahmed El Kholy, Faisal Ahmed, Zhe Gan, Yu Cheng, and Jingjing Liu. 2020. UNITER: UNiversal Image-TExt Representation Learning. arXiv:1909.11740 [cs.CV] https://arxiv.org/abs/1909.11740

    2020 arXiv

  4. [12]

    Bruce Croft

    Van Dang and W. Bruce Croft. 2012. Diversity by proportionality: an election- based approach to search result diversification. InProceedings of the 35th Interna- tional ACM SIGIR Conference on Research and Development in Information Retrieval (Portland, Oregon, USA)(SIGIR ’12)...

    2012

  5. [13]

    Abhimanyu Das and David Kempe. 2011. Submodular meets Spectral: Greedy Algorithms for Subset Selection, Sparse Approximation and Dictionary Selection. arXiv:1102.3975 [stat.ML] https://arxiv.org/abs/1102.3975

    2011 arXiv

  6. [14]

    Shaddin Dughmi. 2011. Submodular Functions: Extensions, Distributions, and Algorithms. A Survey. arXiv:0912.0322 [cs.DS] https://arxiv.org/abs/0912.0322

    2011 arXiv

  7. [15]

    Fleet, Jamie Ryan Kiros, and Sanja Fidler

    Fartash Faghri, David J. Fleet, Jamie Ryan Kiros, and Sanja Fidler

  8. [16]

    Dan Friedman and Adji Bousso Dieng. 2023. The Vendi Score: A Diversity Evaluation Metric for Machine Learning. arXiv:2210.02410 [cs.LG] https://arxiv. org/abs/2210.02410

    2023 arXiv

  9. [17]

    Corrado, Jonathon Shlens, Samy Bengio, Jeffrey Dean, Marc’Aurelio Ranzato, and Tomas Mikolov

    Andrea Frome, Greg S. Corrado, Jonathon Shlens, Samy Bengio, Jeffrey Dean, Marc’Aurelio Ranzato, and Tomas Mikolov. 2013. DeViSE: a deep visual-semantic embedding model. InProceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2(Lake...

    2013

  10. [18]

    Rishabh Iyer and Jeff Bilmes. 2012. Algorithms for approximate minimization of the difference between submodular functions, with applications.arXiv preprint arXiv:1207.0560(2012)

    2012 arXiv

  11. [19]

    Rishabh Iyer and Jeff Bilmes. 2013. Submodular Optimization with Submodular Cover and Submodular Knapsack Constraints. arXiv:1311.2106 [cs.DS] https: //arxiv.org/abs/1311.2106

    2013 arXiv

  12. [20]

    Le, Yunhsuan Sung, Zhen Li, and Tom Duerig

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, and Tom Duerig. 2021. Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision. arXiv:2102.05918 [cs.CV] https://arxiv.org/abs/2102.05918

    2021 arXiv

  13. [21]

    Wonjae Kim, Bokyung Son, and Ildoo Kim. 2021. ViLT: Vision- and-Language Transformer Without Convolution or Region Supervision. arXiv:2102.03334 [stat.ML] https://arxiv.org/abs/2102.03334

    2021 arXiv

  14. [22]

    Krause, C

    A. Krause, C. Guestrin, A. Gupta, and J. Kleinberg. 2006. Near-optimal sensor placements: maximizing information while minimizing communication cost. In 2006 5th International Conference on Information Processing in Sensor Networks. 2–10. doi:10.1145/1127777.1127782

    2006

  15. [23]

    Shamma, Michael S

    Ranjay Krishna, Yuke Zhu, Oliver Groth, Justin Johnson, Kenji Hata, Joshua Kravitz, Stephanie Chen, Yannis Kalantidis, Li-Jia Li, David A. Shamma, Michael S. Bernstein, and Fei-Fei Li. 2016. Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotatio...

    2016 arXiv

  16. [24]

    Alex Kulesza and Ben Taskar. 2011. k-DPPs: fixed-size determinantal point processes. InProceedings of the 28th International Conference on International Con- ference on Machine Learning(Bellevue, Washington, USA)(ICML’11). Omnipress, Madison, WI, USA, 1193–1200

    2011

  17. [25]

    Alex Kulesza and Ben Taskar. 2012. Determinantal Point Processes for Machine Learning.Foundations and Trends®in Machine Learning5, 2–3 (Dec. 2012), 123–286. doi:10.1561/2200000044

    2012 doi

  18. [26]

    Frédéric Lavancier, Jesper Møller, and Ege Rubak. 2014. Determinantal Point Process Models and Statistical Inference.Journal of the Royal Statistical Society Series B: Statistical Methodology77, 4 (Dec. 2014), 853–877. doi:10.1111/rssb.12096

    2014 doi

  19. [27]

    Kuang-Huei Lee, Xi Chen, Gang Hua, Houdong Hu, and Xiaodong He. 2018. Stacked Cross Attention for Image-Text Matching. arXiv:1803.08024 [cs.CV] https://arxiv.org/abs/1803.08024

    2018 arXiv

  20. [28]

    Jure Leskovec, Andreas Krause, Carlos Guestrin, Christos Faloutsos, Jeanne VanBriesen, and Natalie Glance. 2007. Cost-effective outbreak detection in networks. InProceedings of the 13th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining(San Jose, Califo...

    2007

  21. [29]

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. BLIP-2: Boot- strapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models. arXiv:2301.12597 [cs.CV] https://arxiv.org/abs/2301.12597

    2023 arXiv

  22. [30]

    Junnan Li, Ramprasaath Selvaraju, Akhilesh Gotmare, Shafiq Joty, Caim- ing Xiong, and Steven Chu Hong Hoi. 2021. Align before Fuse: Vision and Language Representation Learning with Momentum Distillation. InAd- vances in Neural Information Processing Systems, M. Ranzato, A. Bey...

    2021

  23. [31]

    Xiujun Li, Xi Yin, Chunyuan Li, Pengchuan Zhang, Xiaowei Hu, Lei Zhang, Lijuan Wang, Houdong Hu, Li Dong, Furu Wei, Yejin Choi, and Jianfeng Gao

  24. [32]

    Hui Lin and Jeff Bilmes. 2011. A class of submodular functions for document summarization. InProceedings of the 49th annual meeting of the association for computational linguistics: human language technologies. 510–520

    2011

  25. [33]

    Lawrence Zitnick, and Piotr Dollár

    Tsung-Yi Lin, Michael Maire, Serge Belongie, Lubomir Bourdev, Ross Girshick, James Hays, Pietro Perona, Deva Ramanan, C. Lawrence Zitnick, and Piotr Dollár

  26. [34]

    Russell Lyons. 2003. Determinantal probability measures. arXiv:math/0204325 [math.PR] https://arxiv.org/abs/math/0204325

    2003 arXiv

  27. [35]

    Zelda Mariet, Yaniv Ovadia, and Jasper Snoek. 2019. DPPNet: Approximating Determinantal Point Processes with Deep Networks. arXiv:1901.02051 [stat.ML] https://arxiv.org/abs/1901.02051

    2019 arXiv

  28. [36]

    Baharan Mirzasoleiman, Ashwinkumar Badanidiyuru, Amin Karbasi, Jan Von- drák, and Andreas Krause. 2015. Lazier than lazy greedy. InProceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence(Austin, Texas)(AAAI’15). AAAI Press, 1812–1818

    2015

  29. [37]

    George L Nemhauser, Laurence A Wolsey, and Marshall L Fisher. 1978. An analysis of approximations for maximizing submodular set functions—I.Mathematical programming14, 1 (1978), 265–294

    1978

  30. [38]

    Plummer, Liwei Wang, Chris M

    Bryan A. Plummer, Liwei Wang, Chris M. Cervantes, Juan C. Caicedo, Ju- lia Hockenmaier, and Svetlana Lazebnik. 2016. Flickr30k Entities: Collect- ing Region-to-Phrase Correspondences for Richer Image-to-Sentence Models. arXiv:1505.04870 [cs.CV] https://arxiv.org/abs/1505.04870

    2016 arXiv

  31. [39]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. arXiv:2103.000...

    2021 arXiv

  32. [40]

    Santos, Craig Macdonald, and Iadh Ounis

    Rodrygo L.T. Santos, Craig Macdonald, and Iadh Ounis. 2010. Exploiting query reformulations for web search result diversification. InProceedings of the 19th International Conference on World Wide Web(Raleigh, North Carolina, USA) (WWW ’10). Association for Computing Machinery,...

    2010

  33. [41]

    Rodrygo L. T. Santos, Craig Macdonald, and Iadh Ounis. 2015. Search Result Diversification.Found. Trends Inf. Retr.9, 1 (March 2015), 1–90. doi:10.1561/ 1500000040 MM ’26, November 10–14, 2026, Rio de Janeiro, Brazil Aaryan Sharma, Vishak Prasad C, Virendra Singh, and Ganesh R...

    2015

  34. [42]

    Rodrygo L. T. Santos, Jie Peng, Craig Macdonald, and Iadh Ounis. 2010. Explicit Search Result Diversification through Sub-queries. InAdvances in Information Retrieval, Cathal Gurrin, Yulan He, Gabriella Kazai, Udo Kruschwitz, Suzanne Little, Thomas Roelleke, Stefan Rüger, and ...

    2010

  35. [43]

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki

  36. [44]

    Vasu Singla, Kaiyu Yue, Sukriti Paul, Reza Shirkavand, Mayuka Jayawardhana, Alireza Ganjdanesh, Heng Huang, Abhinav Bhatele, Gowthami Somepalli, and Tom Goldstein. 2024. From Pixels to Prose: A Large Dataset of Dense Image Captions. arXiv:2406.10328 [cs.CV] https://arxiv.org/a...

    2024 arXiv

  37. [45]

    Naoya Sogi, Takashi Shibata, Makoto Terao, Masanori Suganuma, and Takayuki Okatani. 2025. MS-DPPs: Multi-Source Determinantal Point Processes for Con- textual Diversity Refinement of Composite Attributes in Text to Image Retrieval. arXiv:2507.06654 [cs.CV] https://arxiv.org/ab...

    2025 arXiv

  38. [46]

    Zhecheng Wang, Rajanie Prabha, Tianyuan Huang, Jiajun Wu, and Ram Rajagopal

  39. [47]

    Papadopoulos, Agata Lapedriza, Ferda Ofli, Muhammad Imran, and Antonio Torralba

    Ethan Weber, Dim P. Papadopoulos, Agata Lapedriza, Ferda Ofli, Muhammad Imran, and Antonio Torralba. 2022. Incidents1M: a large-scale dataset of images with natural disasters, damage, and incidents. arXiv:2201.04236 [cs.CV] https: //arxiv.org/abs/2201.04236

    2022 arXiv

  40. [48]

    Kai Wei, Rishabh Iyer, and Jeff Bilmes. 2015. Submodularity in Data Subset Selection and Active Learning. InProceedings of the 32nd International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 37), Francis Bach and David Blei (Eds.). PMLR, Lille...

    2015

  41. [49]

    Haolun Wu, Yansen Zhang, Chen Ma, Fuyuan Lyu, Bowei He, Bhaskar Mitra, and Xue Liu. 2024. Result Diversification in Search and Recommendation: A Survey. arXiv:2212.14464 [cs.IR] https://arxiv.org/abs/2212.14464

    2024 arXiv

  42. [50]

    Jie Xu, Dmitri V Kalashnikov, and Sharad Mehrotra. 2014. Efficient summarization framework for multi-attribute uncertain data. InProceedings of the 2014 ACM SIGMOD international conference on Management of data. 421–432

    2014

  43. [51]

    Suorong Yang, Peng Ye, Wanli Ouyang, Dongzhan Zhou, and Furao Shen. 2025. A CLIP-Powered Framework for Robust and Generalizable Data Selection. In The Thirteenth International Conference on Learning Representations. https: //openreview.net/forum?id=9bMZ29SPVx

    2025

  44. [52]

    Ofer Yehuda, Avihu Dekel, Guy Hacohen, and Daphna Weinshall. 2022. Active learning through a covering lens.Advances in Neural Information Processing Systems35 (2022), 22354–22367

    2022

  45. [53]

    Jiahui Yu, Zirui Wang, Vijay Vasudevan, Legg Yeung, Mojtaba Seyedhosseini, and Yonghui Wu. 2022. CoCa: Contrastive Captioners are Image-Text Foundation Models. arXiv:2205.01917 [cs.CV] https://arxiv.org/abs/2205.01917

    2022 arXiv

  46. [54]

    Lu Yuan, Dongdong Chen, Yi-Ling Chen, Noel Codella, Xiyang Dai, Jianfeng Gao, Houdong Hu, Xuedong Huang, Boxin Li, Chunyuan Li, Ce Liu, Mengchen Liu, Zicheng Liu, Yumao Lu, Yu Shi, Lijuan Wang, Jianfeng Wang, Bin Xiao, Zhen Xiao, Jianwei Yang, Michael Zeng, Luowei Zhou, and Pe...

    2021 arXiv

  47. [55]

    Cheng Zhang, Hedvig Kjellstrom, and Stephan Mandt. 2017. Determinantal Point Processes for Mini-Batch Diversification. arXiv:1705.00607 [cs.LG] https: //arxiv.org/abs/1705.00607

    2017 arXiv

  48. [56]

    Pengchuan Zhang, Xiujun Li, Xiaowei Hu, Jianwei Yang, Lei Zhang, Lijuan Wang, Yejin Choi, and Jianfeng Gao. 2021. VinVL: Revisiting Visual Representations in Vision-Language Models. arXiv:2101.00529 [cs.CV] https://arxiv.org/abs/2101. 00529 MASCOT: Model-Aware Submodular Cover...

    2021 arXiv

  49. [2011]

    arXiv:1105.4255 [cs.IR] https://arxiv.org/abs/1105.4255

    Efficient Diversification of Web Search Results. arXiv:1105.4255 [cs.IR] https://arxiv.org/abs/1105.4255

    arXiv

  50. [2015]

    arXiv:1405.0312 [cs.CV] https://arxiv.org/abs/1405.0312

    Microsoft COCO: Common Objects in Context. arXiv:1405.0312 [cs.CV] https://arxiv.org/abs/1405.0312

    arXiv

  51. [2018]

    arXiv:1707.05612 [cs.LG] https://arxiv.org/abs/1707.05612

    VSE++: Improving Visual-Semantic Embeddings with Hard Negatives. arXiv:1707.05612 [cs.LG] https://arxiv.org/abs/1707.05612

    arXiv

  52. [2020]

    arXiv:2004.06165 [cs.CV] https://arxiv.org/abs/2004.06165

    Oscar: Object-Semantics Aligned Pre-training for Vision-Language Tasks. arXiv:2004.06165 [cs.CV] https://arxiv.org/abs/2004.06165

    2004 arXiv

  53. [2021]

    arXiv:2111.02114 [cs.CV] https://arxiv.org/abs/2111.02114

    LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs. arXiv:2111.02114 [cs.CV] https://arxiv.org/abs/2111.02114

    arXiv

  54. [2023]

    arXiv:2312.12856 [cs.CV] https://arxiv.org/abs/2312.12856

    SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing. arXiv:2312.12856 [cs.CV] https://arxiv.org/abs/2312.12856

    arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.