Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Understanding Bias in Perceiving Dimensionality Reduction Projections

T0 review · 4 major / 4 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 Visual interestingness biases practitioners' selection of dimensionality-reduction projections over faithfulness, and color-encoded labels intensify the bias.

desk verdict Creative adversarial study design, but the synthetic faithfulness scores need a manipulation check before the 'over faithfulness' claim holds. read the letter →

arxiv 2507.20805 v1 pith:YKHF2OVN submitted 2025-07-28 cs.HC cs.LG

Seoyoung Doh , Hyeon Jeon , Sungbok Shin , Ghulam Jilani Quadri , Nam Wook Kim , Jinwook Seo This is my paper · ORCID
classification cs.HCcs.LG
keywords visualinterestingnessdimensionalityreductionfaithfulnessperceptualbiasdual-systemtheoryuserstudyscatterplotperceptionanalyticalpreference
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 sets out to show that when practitioners choose a dimensionality-reduction (DR) projection for analysis, they are systematically pulled by how visually interesting the scatterplot looks rather than by how faithfully it preserves the data structure. In a two-phase study with separate groups, one phase ranked projections by visual appeal and the other ranked the same projections by analytical preference while showing faithfulness scores deliberately rigged to favor the less appealing projections. Analytical-preference rankings still tracked visual appeal, with a positive Spearman correlation, and the effect grew significantly stronger when class labels were color-coded. The paper also hypothesized that shorter viewing time would intensify the bias, but the measured effect was not significant; interviews suggested weak support. If the claim holds, common practices for evaluating DR outputs can be undermined by a perceptual bias that users are not aware of.

What carries the argument

The experimental machinery has three load-bearing parts. First, an active ranking algorithm converts pairwise "which is more visually interesting?" choices into stable rankings of 20 projections per condition with minimal trials. Second, the adversarial faithfulness-score generator assigns random scores in [0,1] so that the less interesting projection wins on three or four of five metrics, creating a direct conflict between what looks good and what the numbers say. Third, Spearman rank correlations between the visual-interest rankings and the analytical-preference rankings, analyzed with two-way ANOVA, quantify the bias; dual-system theory supplies the explanatory frame, mapping visual appeal to System 1 and score-based evaluation to System 2.

What would settle it

A replication that replaces the synthetic scores with real, verified faithfulness metrics for the same projections and tells participants the scores are trustworthy; if preferences then track faithfulness rather than visual interest, the original result depended on distrust of the numbers, not on visual bias. A cheaper check is to add a post-task question asking participants how much they believed the synthetic scores predicted true quality.

Watch

Extended reading notes

Core claim

The central discovery is that visual interestingness—the degree to which a projection shows salient, distinctive, or aesthetically appealing patterns—dominates analytical preference for DR projections even when faithfulness scores explicitly favor a different projection. The authors created pairwise comparisons where five synthetic metrics, labeled A through E, were randomly generated so that three or four of them gave higher scores to the less visually interesting projection. Across 16 combinations of color encoding and exposure time, the Spearman correlation between visual-interest rankings and analytical-preference rankings stayed positive (around 0.25), and a two-way ANOVA found a significant effect of color encoding (F1,336 = 65.10, p < .001) but no significant effect of exposure time. Interviews confirmed that participants relied on visual appearance, felt that color-coded projections drew more attention, and were largely unaware of the bias. The paper interprets this through dual-system theory: visual interest triggers fast System 1 processing, while faithfulness requires slower System 2 reasoning that is often skipped.

Load-bearing premise

In Phase 2, the claim rests on participants accepting five randomly generated numbers, labeled only "metrics A through E" and rigged to favor less interesting projections, as credible indicators of true faithfulness; if they treated those numbers as meaningless or fake, the experiment would show a preference for attractive plots over arbitrary numbers, not over actual faithfulness.

Editorial extensions

If this is right

  • When faithfulness scores conflict with visual appeal, practitioners tend to follow the visual appeal, so projection-selection interfaces that only show metric values may fail to steer analysts toward faithful embeddings.
  • Color-encoding class labels amplifies the bias, so monochrome displays or shape encodings are plausible mitigations, as the paper recommends.
  • Merely giving analysts more time may not remove the bias, since participants in the study decided in about five seconds even when allowed fifteen.
  • Projections with well-separated classes and clumped clusters are perceived as visually interesting, so these visual features are the ones most likely to trigger the bias.
  • The paper's proposed mitigation strategies—deactivating System 1 through visual simplification, making faithfulness scores visually salient, and improving DR literacy—are direct consequences of the identified mechanism and remain to be evaluated.

Reading between the lines

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

  • The same perceptual mechanism likely extends beyond DR projection choice to any visual analytics setting where a salient, attractive layout competes with an objective quality score, so the bias may be a general visualization phenomenon rather than DR-specific.
  • The adversarial-score design could be reused as a calibration instrument: presenting the same projections with real, verified faithfulness metrics would test whether the bias persists when the numbers are known to be trustworthy.
  • Because participants were unaware of the bias and some denied it, a practical next step is to measure whether simply warning analysts about the bias, or requiring them to justify their choice, changes preferences.
  • Individual-level analyses could test whether DR literacy moderates the effect; the paper reports bias across self-reported literacy levels but does not statistically compare subgroups.
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 / 4 minor

Summary. The paper reports a two-phase user study testing whether practitioners' analytical preference for dimensionality reduction (DR) projections is biased toward visual interestingness over faithfulness. In Phase 1, 16 participants produced visual-interest rankings of projections via active pairwise comparisons; in Phase 2, 16 different participants chose projections for cluster analysis while viewing five artificially generated, adversarially assigned 'faithfulness' scores. The authors test three hypotheses: H1 that visual interestingness more strongly influences analytical preference than faithfulness, H2 that color encoding intensifies the bias, and H3 that shorter exposure time intensifies the bias. The quantitative result is a significant effect of color encoding on the correlation between visual-interest and analytical-preference rankings (F1,336 = 65.10, p < .001) and no significant effect of exposure time (F1,336 = 0.22, p = 0.64). The paper also reports a regression analysis of visual features and qualitative interview findings, and proposes mitigation strategies.

Significance. If the central claim is upheld after revision, the paper addresses an important and timely problem for visualization and visual analytics: users may choose DR projections for perceptual appeal rather than structural fidelity. The adversarial score assignment is a genuinely useful design because it makes a null result possible, and the positive correlation plus the color-encoding effect are promising evidence. The qualitative findings on bias awareness and the proposed mitigation strategies are also valuable. However, the central inference depends on participants accepting fabricated faithfulness scores as credible, which is not checked, and the paper currently overstates H3 in the abstract. These issues are fixable within the manuscript's scope and do not require rejecting the work.

major comments (4)
  1. [Abstract and §4.1] The abstract states that the bias 'intensifies with color-encoded labels and shorter exposure time,' but the quantitative analysis in §4.1 reports no significant effect of EXPOSURE TIME (F1,336 = 0.22, p = 0.64) and no interaction. The qualitative Finding 2 in §5, based on ten self-reports, is explicitly called 'weak support' by the authors. The shorter-exposure claim must be removed from the abstract or explicitly qualified as qualitative-only, and the discussion should reconcile the quantitative null result with the qualitative reports.
  2. [§3.2.2 (Generating and presenting faithfulness scores)] The Phase 2 manipulation relies on participants treating five arbitrary numbers in [0,1], labeled only 'metrics A through E' and rigged to favor less visually interesting projections, as credible indicators of projection faithfulness. The paper provides no manipulation check, no comprehension check, and no pilot evidence that participants understood these scores as genuine faithfulness information. Without such a check, the positive correlation between visual-interest rankings and analytical-preference rankings could reflect a preference for visually interesting plots over arbitrary numeric displays rather than a bias against faithfulness. This is load-bearing for H1, so the authors should either add a manipulation check or a post-task credibility rating, or substantially reframe the claim to avoid equating the synthetic scores with faithfulness.
  3. [§4.1 (Analysis design)] The statistical reporting is internally inconsistent. The text says there are 16 combinations (2 color × 2 exposure × 4 dataset variations), each contributing 4 × 4 = 16 Spearman correlations, which yields 256 observations, but the reported ANOVA degrees of freedom (F1,336) imply 340 observations after four model parameters. Please clarify the exact unit of analysis, report how Phase 1 and Phase 2 rankings were paired despite having different participants, and re-run the analysis at the correct unit or explain the discrepancy.
  4. [§4.1 (Results and discussions)] The paper reports only that correlations 'range around 0.25' and gives F and p values for the ANOVA, but it never reports the mean, standard deviation, or confidence intervals for the Spearman correlations in each condition, nor the effect size for the color effect. Since the central conclusion is that the correlation is positive and larger with color encoding, the authors should report these descriptive statistics and effect sizes explicitly, including the values shown only in Figure 3.
minor comments (4)
  1. [§4.1] The sentence 'We find a significant effect on COLOR ENCODING ... confirming H1' appears to conflate H1 with H2; the color effect supports H2, while H1 is supported by the positive correlation itself. Please reword to avoid confusing the reader.
  2. [§3.2.1 (Stimuli)] The stimuli-sampling procedure is deferred to 'Appendix A,' but no appendix is present in the submitted manuscript. Please include the appendix or add a brief in-text description so the stratified sampling of datasets and projections is reproducible.
  3. [§5 (Finding 2)] The qualitative evidence for the exposure-time effect is based on self-reports in an interview setting where participants may be primed by the study design; the text already calls this 'weak support,' but the authors should avoid using it to endorse H3 in the conclusion and abstract.
  4. [§3.2.2] Please clarify whether the 'five pairs of faithfulness scores' means five scores per projection or five paired comparisons per trial; the current wording is ambiguous and the reader cannot determine how the scores were visually arranged for the participant.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the bias claim rests on separately measured rankings and an adversarial score assignment that permits a null result.

full rationale

The paper is an empirical user study rather than a derivation. Visual-interest rankings (Phase 1) and analytical-preference rankings (Phase 2) are elicited from different participants with different questions, and the faithfulness scores are artificially generated to favor the less interesting projection ('we assign higher scores to the less visually interesting projection on three or four randomly selected metrics'). This adversarial construction makes a negative or zero correlation possible, so the observed positive Spearman correlation is not forced by construction. The central inference therefore does not reduce to its inputs. Authors' prior work is cited for auxiliary materials (the 96-dataset pool [13], the ZADU metric library [14], CLAMS cluster-count estimation [17]) and for discussion of t-SNE/UMAP misuse [16,15]; these are external published artifacts used as tools or background, not as the evidence establishing the bias, so they do not constitute load-bearing self-citation. The only substantive threat—that participants may not have treated the fake A–E scores as credible faithfulness information—is a manipulation-check/construct-validity concern, not a circularity, because the outcome measure was not defined in terms of, nor fitted to, the predictor. No equation or fitted parameter is renamed as a prediction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central behavioral claim rests on several domain assumptions rather than derivations. The most important are that the synthetic metrics are perceived as real faithfulness information, that the active ranking recovers stable rankings, and that the student sample represents practitioners. No free parameters are fitted to data in the traditional sense; the design parameters listed are hand-chosen study settings that affect the claims.

free parameters (3)
  • Faithfulness score generation rule = 3 or 4 of 5 metrics favor the less interesting projection; values uniform in [0,1]
    Chosen by the authors to force a conflict between visual interest and faithfulness; if participants do not perceive these scores as valid, the central tradeoff measurement is undermined.
  • Visual interestingness score transformation = score = 21 - rank
    Used to turn consensus rankings into a numeric target for linear regression in Analysis 2; the arbitrary linear scaling affects R2 values but not the ranking of feature deletions.
  • Exposure time levels = 7 s and 15 s
    Selected from a pilot; the manipulation did not work because participants responded in about five seconds, so the shorter exposure time hypothesis was not actually tested.
assumptions (4)
  • domain assumption Dual-system theory (System 1 and System 2) describes the cognitive processes at play in DR projection selection.
    The paper maps visual interestingness to System 1 and faithfulness to System 2 (Sect. 2.2) but never measures these systems directly; the explanation for the bias depends on this imported theory.
  • domain assumption Artificial metrics A-E are accepted by participants as faithful indicators of projection quality.
    In Phase 2 (Sect. 3.2.2), random scores labeled generically are assumed to carry the meaning of faithfulness; no manipulation check or debrief detail is provided.
  • domain assumption The active ranking algorithm produces stable and valid rankings from 50 pairwise comparisons per 20 projections.
    All ranking analyses in Sect. 4 rely on the algorithm's output as ground truth, but the paper reports no reliability check, such as test-retest or synthetic validation.
  • domain assumption University students with scatterplot experience are representative of practitioners.
    Participants are recruited from local universities (Sect. 3.2.1), yet the paper generalizes to practitioners.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Understanding Bias in Perceiving Dimensionality Reduction Projections." pith.science (2026). https://pith.science/paper/YKHF2OVN

@misc{pith2026250720805,
  author       = {Pith},
  title        = {Pith review of: Understanding Bias in Perceiving Dimensionality Reduction Projections},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YKHF2OVN}},
  note         = {Machine review of arXiv:2507.20805}
}
read the original abstract

Selecting the dimensionality reduction technique that faithfully represents the structure is essential for reliable visual communication and analytics. In reality, however, practitioners favor projections for other attractions, such as aesthetics and visual saliency, over the projection's structural faithfulness, a bias we define as visual interestingness. In this research, we conduct a user study that (1) verifies the existence of such bias and (2) explains why the bias exists. Our study suggests that visual interestingness biases practitioners' preferences when selecting projections for analysis, and this bias intensifies with color-encoded labels and shorter exposure time. Based on our findings, we discuss strategies to mitigate bias in perceiving and interpreting DR projections.

Figures

Figures reproduced from arXiv: 2507.20805 by the authors.

Figure 1
Figure 1. The illustration of our motivating scenario and its alignment with dual-system theory ( [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The illustration of our study design (Sect. 3). In Phase 1, participants view pairs of projections and select the one with higher visual interestingness. Phase 2 repeats this process for analytical preference, but with artificially generated faithfulness metrics that favor less visually interesting projections. This design enables us to quantify participants’ bias toward visual interestingness when it conflicts with… view at source ↗
Figure 3
Figure 3. The correlations between visual interestingness and analyt [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The time duration needed to select projections in our exper [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Stop Misusing t-SNE and UMAP for Visual Analytics

    cs.HC 2025-06 conditional novelty 5.0 of 10

    A review of 136 visual analytics papers and two interview studies show t-SNE and UMAP are frequently used for tasks they cannot support, and that practitioner literacy issues explain the persistent misuse.

  2. FlexMUSE: Multimodal Unification and Semantics Enhancement Framework with Flexible interaction for Creative Writing

    cs.CV 2025-08 reject novelty 4.0 of 10

    FlexMUSE, a claimed multimodal creative-writing framework and its ArtMUSE dataset, are unsupported because the submitted full text is an unrelated dimensionality-reduction paper (UMATO).

Reference graph

Works this paper leans on

39 extracted references · 21 canonical work pages · cited by 2 Pith papers

  1. [1]

    M. M. Abbas, M. Aupetit, M. Sedlmair, and H. Bensmail. Clustme: A visual quality measure for ranking monochrome scatterplots based on cluster patterns. Computer Graphics Forum, 38(3):225–236, 2019. doi: 10.1111/cgf.13684 3

    doi:10.1111/cgf.13684 2019

  2. [2]

    S. S. Bae, T. Fujiwara, C. Tseng, and D. Szafir. Uncovering how scat- terplot features skew visual class separation. In ACM CHI, 2025. 4

    work page 2025

  3. [3]

    Bernard, M

    J. Bernard, M. Hutter, M. Zeppelzauer, D. Fellner, and M. Sedlmair. Comparing visual-interactive labeling with active learning: An exper- imental study. IEEE Transactions on Visualization and Computer Graphics, 24(1):298–308, 2018. doi: 10.1109/TVCG.2017.2744818 5

    work page arXiv 2018

  4. [4]

    Bibal and B

    A. Bibal and B. Fr ´enay. Learning interpretability for visualizations using adapted cox models through a user experiment, 2016. 2, 4

    work page 2016

  5. [5]

    Cashman, M

    D. Cashman, M. Keller, H. Jeon, B. C. Kwon, and Q. Wang. A criti- cal analysis of the usage of dimensionality reduction in four domains. IEEE Transactions on Visualization and Computer Graphics , pp. 1– 20, 2025. doi: 10.1109/TVCG.2025.3567989 1, 2

    work page arXiv 2025

  6. [6]

    Colin Cameron and F

    A. Colin Cameron and F. A. Windmeijer. An r-squared measure of goodness of fit for some common nonlinear regression models. Jour- nal of Econometrics, 77(2):329–342, 1997. doi:10.1016/S0304-4076(96) 01818-0 4

    work page doi:10.1016/s0304-4076(96 1997

  7. [7]

    Correll and M

    M. Correll and M. Gleicher. Error bars considered harmful: Exploring alternate encodings for mean and error. IEEE Transactions on Visu- alization and Computer Graphics , 20(12):2142–2151, 2014. doi: 10. 1109/TVCG.2014.2346298 5

    arXiv 2014

  8. [8]

    J. H. Friedman. Exploratory projection pursuit. Journal of the Ameri- can statistical association, 82(397):249–266, 1987. 1, 2

    work page 1987

Show all 39 references
  1. [9]

    Goffin, J

    P. Goffin, J. Boy, W. Willett, and P. Isenberg. An exploratory study of word-scale graphics in data-rich text documents. IEEE Transactions on Visualization and Computer Graphics , 23(10):2275–2287, 2017. doi: 10.1109/TVCG.2016.2618797 5

    2017

  2. [10]

    T. M. Green, W. Ribarsky, and B. Fisher. Visual analytics for complex concepts using a human cognition model. In 2008 IEEE Symposium on Visual Analytics Science and Technology, pp. 91–98, 2008. doi: 10. 1109/VAST.2008.4677361 2, 5

    2008

  3. [11]

    Healey and J

    C. Healey and J. Enns. Attention and visual memory in visualization and computer graphics. IEEE Transactions on Visualization and Com- puter Graphics, 18(7):1170–1188, 2012. doi: 10.1109/TVCG.2011.127 2, 3

    2012 doi

  4. [12]

    K. G. Jamieson and R. Nowak. Active ranking using pairwise compar- isons. Advances in neural information processing systems , 24, 2011. 3, 4

    2011

  5. [13]

    H. Jeon, M. Aupetit, D. Shin, A. Cho, S. Park, and J. Seo. Measuring the validity of clustering validation datasets. IEEE Transactions on Pattern Analysis and Machine Intelligence , 47(6):5045–5058, 2025. doi: 10.1109/TPAMI.2025.3548011 3

    2025

  6. [14]

    H. Jeon, A. Cho, J. Jang, S. Lee, J. Hyun, H.-K. Ko, J. Jo, and J. Seo. Zadu: A python library for evaluating the reliability of dimensionality reduction embeddings. In 2023 IEEE Visualization and Visual Ana- lytics (VIS), pp. 196–200. doi: 10.1109/VIS54172.2023.00048 2, 4

    2023

  7. [15]

    H. Jeon, H. Lee, Y .-H. Kuo, T. Yang, D. Archambault, S. Ko, T. Fuji- wara, K.-L. Ma, and J. Seo. Unveiling high-dimensional backstage: A survey for reliable visual analytics with dimensionality reduction. In Proceedings of the 2025 CHI Conference on Human Factors in Com- puti...

    2025

  8. [16]

    H. Jeon, J. Park, S. Shin, and J. Seo. Stop misusing t-sne and umap for visual analytics, 2025. 1, 2, 5

    2025

  9. [17]

    H. Jeon, G. J. Quadri, H. Lee, P. Rosen, D. A. Szafir, and J. Seo. Clams: A cluster ambiguity measure for estimating perceptual vari- ability in visual clustering. IEEE Transactions on Visualization and Computer Graphics, 30(1):770–780, 2024. doi: 10.1109/TVCG.2023 .3327201 2, 3, 4

    2024 doi

  10. [18]

    P. Joia, D. Coimbra, J. A. Cuminato, F. V . Paulovich, and L. G. Nonato. Local affine multidimensional projection. IEEE Transactions on Vi- sualization and Computer Graphics, 17(12):2563–2571, 2011. doi: 10 .1109/TVCG.2011.220 4

    2011

  11. [19]

    Kahneman

    D. Kahneman. Thinking, fast and slow. 2011. 2

    2011

  12. [20]

    Kobak and G

    D. Kobak and G. C. Linderman. Initialization is critical for preserving global data structure in both t-sne and umap. Nature biotechnology, 39(2):156–157, 2021. doi: 10.1038/s41587-020-00809-z5

    2021 doi

  13. [21]

    B. C. Kwon, B. Eysenbach, J. Verma, K. Ng, C. De Filippi, W. F. Stewart, and A. Perer. Clustervision: Visual supervision of unsuper- vised clustering. IEEE Transactions on Visualization and Computer Graphics, 24(1):142–151, 2018. doi: 10.1109/TVCG.2017.2745085 2, 3, 4

    2018

  14. [22]

    L. v. d. Maaten and G. Hinton. Visualizing data using t-sne. Journal of machine learning research, 9(Nov):2579–2605, 2008. 5

    2008

  15. [23]

    McInnes, J

    L. McInnes, J. Healy, and J. Melville. Umap: Uniform manifold ap- proximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426, 2018. 5

    2018 arXiv

  16. [24]

    Morariu, A

    C. Morariu, A. Bibal, R. Cutura, B. Fr ´enay, and M. Sedlmair. Pre- dicting user preferences of dimensionality reduction embedding qual- ity. IEEE Transactions on Visualization and Computer Graphics , 29(1):745–755, 2023. doi: 10.1109/TVCG.2022.3209449 2, 4

    2023

  17. [25]

    Nguyen, P

    Q. Nguyen, P. Eades, and S.-H. Hong. On the faithfulness of graph visualizations. In 2013 IEEE Pacific Visualization Symposium (Paci- ficVis), pp. 209–216, 2013. doi: 10.1109/PacificVis.2013.6596147 1, 2

    2013

  18. [26]

    L. G. Nonato and M. Aupetit. Multidimensional projection for visual analytics: Linking techniques with distortions, tasks, and layout en- richment. IEEE Transactions on Visualization and Computer Graph- ics, 25(8):2650–2673, 2019. doi: 10.1109/TVCG.2018.2846735 1, 2

    2019

  19. [27]

    A. V . Pandey, J. Krause, C. Felix, J. Boy, and E. Bertini. Towards understanding human similarity perception in the analysis of large sets of scatter plots. InProceedings of the 2016 CHI Conference on Human Factors in Computing Systems, CHI ’16, p. 3659–3669, 2016. doi: 10. 1...

    2016

  20. [28]

    Seo and B

    J. Seo and B. Shneiderman. A rank-by-feature framework for interac- tive exploration of multidimensional data. Information visualization, 4(2):96–113, 2005. 1, 2

    2005

  21. [29]

    South, D

    L. South, D. Saffo, O. Vitek, C. Dunne, and M. A. Borkin. Effective use of likert scales in visualization evaluations: A systematic review. Computer Graphics Forum, 41(3):43–55, 2022. doi: 10.1111/cgf.14521 3

    2022 doi

  22. [30]

    Strobelt, D

    H. Strobelt, D. Oelke, B. C. Kwon, T. Schreck, and H. Pfister. Guide- lines for effective usage of text highlighting techniques. IEEE Trans- actions on Visualization and Computer Graphics , 22(1):489–498,

  23. [31]

    Tseng, A

    C. Tseng, A. Z. Wang, G. J. Quadri, and D. A. Szafir. Shape it up: An empirically grounded approach for designing shape palettes. IEEE Transactions on Visualization and Computer Graphics , 31(1):349– 359, 2025. doi: 10.1109/TVCG.2024.3456385 5

    2025

  24. [32]

    Tversky and D

    A. Tversky and D. Kahneman. Judgment under uncertainty: Heuristics and biases: Biases in judgments reveal some heuristics of thinking under uncertainty. science, 185(4157):1124–1131, 1974. 2

    1974

  25. [33]

    A. C. Valdez, M. Ziefle, and M. Sedlmair. Priming and anchoring ef- fects in visualization. IEEE Transactions on Visualization and Com- puter Graphics, 24(1):584–594, 2018. doi: 10.1109/TVCG.2017.2744138 3

    2018

  26. [34]

    Venna and S

    J. Venna and S. Kaski. Local multidimensional scaling. Neural Net- works, 19(6):889–899, 2006. Advances in Self Organising Maps - WSOM’05. doi: 10.1016/j.neunet.2006.05.014 2

    2006 doi

  27. [35]

    Y . Wang, K. Feng, X. Chu, J. Zhang, C.-W. Fu, M. Sedlmair, X. Yu, and B. Chen. A perception-driven approach to supervised dimension- ality reduction for visualization. IEEE Transactions on Visualization and Computer Graphics , 24(5):1828–1840, 2018. doi: 10.1109/TVCG. 2017.27...

    2018

  28. [36]

    Wattenberg, F

    M. Wattenberg, F. Vi´egas, and I. Johnson. How to use t-sne effectively. Distill, 2016. doi: 10.23915/distill.00002 2, 5

    2016 doi

  29. [37]

    Wilkinson, A

    L. Wilkinson, A. Anand, and R. Grossman. Graph-theoretic scagnos- tics. In Information visualization, IEEE symposium on , pp. 21–21. IEEE Computer Society, 2005. 4

    2005

  30. [38]

    J. Xia, Y . Zhang, J. Song, Y . Chen, Y . Wang, and S. Liu. Revisit- ing dimensionality reduction techniques for visual cluster analysis: An empirical study. IEEE Transactions on Visualization and Com- puter Graphics, 28(1):529–539, 2022. doi: 10.1109/TVCG.2021.3114694 2, 3

    2022

  31. [2016]

    doi: 10.1109/TVCG.2015.2467759 5

    2015

Pith tools

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