pith. machine review for the scientific record. sign in

arxiv: 2604.20210 · v2 · submitted 2026-04-22 · 💻 cs.HC · cs.AI· cs.LG

Recognition: unknown

Vibrotactile Preference Learning: Uncertainty-Aware Preference Learning for Personalized Vibration Feedback

Rongtao Zhang , Xin Zhu , Masoume Pourebadi Khotbehsara , Warren Dao , Erdem B{\i}y{\i}k , Heather Culbertson
Authors on Pith no claims yet

Pith reviewed 2026-05-09 23:58 UTC · model grok-4.3

classification 💻 cs.HC cs.AIcs.LG
keywords vibrotactilepreference learningpersonalizationhaptic feedbackgaussian processhuman computer interactionuncertainty awareuser study
0
0 comments X

The pith

Vibrotactile Preference Learning captures each person's unique vibration preferences from pairwise comparisons that include uncertainty reports.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper shows how to personalize the feel of vibrations in devices by learning what each user likes. It does so by asking users to compare two vibrations at a time, note how certain they are about which one they prefer, and then picking the next pair based on what information would help most. A test with thirteen participants using an Xbox controller found that after forty such rounds the system had a good model of their preferences, and people did not find the process tiring. This matters because many devices now use vibration feedback, yet one setting does not suit everyone, so efficient personalization could improve comfort and engagement in games, notifications, and interfaces.

Core claim

VPL captures user-specific preference spaces over vibrotactile parameters via Gaussian-process-based uncertainty-aware preference learning. It uses an expected information gain-based acquisition strategy to guide query selection over 40 rounds of pairwise comparisons of overall user preference, augmented with user-reported uncertainty. This enables efficient exploration of the parameter space. A user study with N=13 using Microsoft Xbox controller feedback shows that VPL efficiently learns individualized preferences while maintaining comfortable, low-workload user interactions.

What carries the argument

Vibrotactile Preference Learning (VPL), a Gaussian process model that updates from uncertainty-augmented pairwise preference queries selected to maximize expected information gain.

If this is right

  • Personalized vibration settings become available for each user after a short calibration session of 40 comparisons.
  • Devices can adapt haptic feedback to individual perception differences without high user effort.
  • Exploration of the vibration parameter space stays efficient because queries focus on high-uncertainty areas.
  • Scalable personalization of haptics in interactive systems is supported by the low-workload design.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar uncertainty-aware methods could apply to learning preferences in other feedback types such as sound or visual effects.
  • Integrating the model into ongoing use might allow preferences to update as users gain experience with the device.
  • Testing across different hardware, like phones or wearables, would check how well the learned preferences transfer.

Load-bearing premise

That users can consistently report their uncertainty about vibration preferences in a way that improves the model's accuracy over just using their choices alone.

What would settle it

Running the system on new users and then checking whether they actually select the model's top-predicted vibration over others in blind tests; if they do not, the learning did not capture true preferences.

Figures

Figures reproduced from arXiv: 2604.20210 by Erdem B{\i}y{\i}k, Heather Culbertson, Masoume Pourebadi Khotbehsara, Rongtao Zhang, Warren Dao, Xin Zhu.

Figure 1
Figure 1. Figure 1: Overview of the Vibrotactile Preference Learning System [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: VPL Pipeline: Users make pairwise vibrotactile judgments and report confidence; a confidence-aware GP preference [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Simulation results 4.2 Participants We recruited 13 participants (P01–P13; 8 female, 5 male) aged 18 to 43 years (𝑀 = 28.8, 𝑆𝐷 = 7.2). Participants reported varying gam￾ing frequencies (Never: 4, Occasionally: 8, Often: 1) and controller familiarity (Low: 3, Medium: 3, High: 7). The study was approved by the University Institutional Review Board, and all participants gave informed consent. 4.3 Experimental… view at source ↗
Figure 4
Figure 4. Figure 4: Validation accuracy grouped by item type [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Validation accuracy grouped by participant [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: NASA-TLX workload ratings. (a) EUBO sampling (b) InfoGain-style sampling [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Illustrative application domains where VPL could support personalized vibrotactile feedback. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
read the original abstract

Individual differences in vibrotactile perception underscore the growing importance of personalization as haptic feedback becomes more prevalent in interactive systems. We propose Vibrotactile Preference Learning (VPL), a system that captures user-specific preference spaces over vibrotactile parameters via Gaussian-process-based uncertainty-aware preference learning. VPL uses an expected information gain-based acquisition strategy to guide query selection over 40 rounds of pairwise comparisons of overall user preference, augmented with user-reported uncertainty, enabling efficient exploration of the parameter space. We evaluate VPL in a user study (N = 13) using the vibrotactile feedback from a Microsoft Xbox controller, showing that it efficiently learns individualized preferences while maintaining comfortable, low-workload user interactions. These results highlight the potential of VPL for scalable personalization of vibrotactile experiences.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The manuscript introduces Vibrotactile Preference Learning (VPL), a Gaussian-process-based framework for capturing user-specific preference spaces over vibrotactile parameters. VPL performs 40 rounds of pairwise preference comparisons augmented by user-reported uncertainty, with queries chosen via expected information gain acquisition. A user study (N=13) using Microsoft Xbox controller vibrotactile feedback reports that the system learns individualized preferences efficiently while keeping interactions comfortable and low-workload.

Significance. If the empirical results are robust, the work is significant for HCI and haptics research. It directly tackles individual differences in vibrotactile perception, an increasingly relevant issue as haptic feedback proliferates in consumer devices. The uncertainty-aware preference model combined with information-theoretic query selection offers a practical path toward scalable personalization that reduces user burden compared with exhaustive sampling. The explicit incorporation of user uncertainty and the focus on workload metrics are strengths that could influence future preference-learning systems in VR, gaming, and wearable interfaces.

major comments (1)
  1. §4 (User Study): The central efficiency claim rests on the N=13 study outcomes, yet the manuscript provides insufficient detail on the statistical tests, effect sizes, or baseline comparisons (e.g., random query selection or non-uncertainty-aware GP). Without these, it is difficult to assess whether the reported gains in preference learning speed and workload reduction are statistically reliable or attributable to the proposed components.
minor comments (3)
  1. §3.2 (Method): The exact form of the Gaussian process kernel and how user uncertainty is encoded as observation noise should be stated explicitly with an equation, rather than described only in prose.
  2. Figure 3 (Results): The learning curves would benefit from error bars or confidence intervals to allow visual assessment of variability across participants.
  3. §5 (Discussion): The manuscript should briefly address potential limitations of the 40-round budget and the Xbox controller hardware when generalizing to other vibrotactile devices.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment of our work's significance for HCI and haptics, and for the constructive feedback on the user study analysis. We address the concern below and will revise the manuscript to incorporate additional statistical details and comparisons.

read point-by-point responses

  1. Referee: §4 (User Study): The central efficiency claim rests on the N=13 study outcomes, yet the manuscript provides insufficient detail on the statistical tests, effect sizes, or baseline comparisons (e.g., random query selection or non-uncertainty-aware GP). Without these, it is difficult to assess whether the reported gains in preference learning speed and workload reduction are statistically reliable or attributable to the proposed components.

    Authors: We agree that the current presentation of results in §4 lacks sufficient statistical rigor and baseline comparisons, which limits interpretability of the efficiency claims. In the revised manuscript, we will add: (1) explicit reporting of statistical tests used (e.g., paired t-tests or non-parametric equivalents for within-subject comparisons of learning curves and workload scores), (2) effect sizes (Cohen's d or equivalent), and (3) baseline comparisons. For baselines, we will include simulated results for random query selection and a non-uncertainty-aware GP model (using the same preference data where possible, or additional offline simulations on the collected preference pairs) to isolate the contributions of uncertainty awareness and expected information gain. These will be integrated into §4 with updated figures/tables, ensuring the gains are shown to be statistically reliable and attributable to the proposed components. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper describes an empirical user study (N=13) that applies standard Gaussian-process preference learning with expected information gain acquisition and user-reported uncertainty over 40 pairwise comparisons. All load-bearing elements are externally validated by the collected preference data and workload metrics rather than reducing to self-definition, fitted inputs renamed as predictions, or self-citation chains. The modeling pipeline follows established GP preference-learning practice without importing uniqueness theorems or ansatzes from the authors' prior work as load-bearing premises. The central claim of efficient individualized learning is therefore supported by independent experimental outcomes.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only; no specific free parameters, axioms, or invented entities are identifiable from the summary. The central modeling choice relies on standard Gaussian process assumptions for preference learning.

axioms (1)
  • domain assumption User preferences over vibrotactile parameters can be modeled as a Gaussian process
    This is the core modeling assumption enabling uncertainty-aware learning.

pith-pipeline@v0.9.0 · 5468 in / 1217 out tokens · 38739 ms · 2026-05-09T23:58:55.493349+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 12 canonical work pages · 1 internal anchor

  1. [1]

    Raul Astudillo and Peter I Frazier. 2023. qEUBO: A Decision-Theoretic Acqui- sition Function for Preferential Bayesian Optimization. InProceedings of the 26th International Conference on Artificial Intelligence and Statistics (AISTATS) (Proceedings of Machine Learning Research)

    2023

  2. [2]

    Mauriello, So Yeon Park, James A

    Stephanie Balters, Matthew L. Mauriello, So Yeon Park, James A. Landay, and Pablo E. Paredes. 2020. Calm Commute: Guided Slow Breathing for Daily Stress Management in Drivers.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies4, 1 (2020), 1–19

    2020

  3. [3]

    Montiel, and Heather Culbertson

    Premankur Banerjee, Jiaxuan Wang, Lauren Tomita, Mia P. Montiel, and Heather Culbertson. 2025. Virtual Encounters of the Haptic Kind: Towards a Multi- User VR System for Real-Time Social Touch. arXiv:2502.13421 [cs.HC] https: //arxiv.org/abs/2502.13421

    work page arXiv 2025

  4. [4]

    Kochenderfer, and Dorsa Sadigh

    Erdem Bıyık, Nicolas Huynh, Mykel J. Kochenderfer, and Dorsa Sadigh. 2024. Active Preference-Based Gaussian Process Regression for Reward Learning and Optimization.The International Journal of Robotics Research43, 5 (2024), 665–684. doi:10.1177/02783649231208729

    work page doi:10.1177/02783649231208729 2024

  5. [5]

    Ralph Allan Bradley and Milton E. Terry. 1952. Rank Analysis of Incomplete Block Designs: I. The Method of Paired Comparisons.Biometrika39, 3–4 (1952), 324–345

    1952

  6. [6]

    A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning

    Eric Brochu, Vlad M. Cora, and Nando de Freitas. 2010. A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning. arXiv:1012.2599

    work page Pith review arXiv 2010

  7. [7]

    Eric Brochu, Nando de Freitas, and Abhijeet Ghosh. 2007. Active Preference Learning with Discrete Choice Data. InAdvances in Neural Information Processing Systems

    2007

  8. [8]

    Picard, and Hiroshi Ishii

    Kyung Yun Choi, Nizer ElHaouij, Jeon Lee, Rosalind W. Picard, and Hiroshi Ishii

  9. [9]

    Design and Evaluation of a Clippable and Personalizable Pneumatic-Haptic Feedback Device for Breathing Guidance.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies6, 1 (2022), 1–36

    2022

  10. [10]

    Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. Deep reinforcement learning from human preferences.Advances in neural information processing systems30 (2017)

    2017

  11. [11]

    Wei Chu and Zoubin Ghahramani. 2005. Preference Learning with Gaussian Processes. InProceedings of the 22nd International Conference on Machine Learning (ICML). 137–144

    2005

  12. [12]

    Foley, David V

    Hugh J. Foley, David V. Cross, and James A. O’Reilly. 1990. Pervasiveness and Mag- nitude of Context Effects: Evidence for the Relativity of Absolute Magnitude Esti- mation.Perception & Psychophysics48, 6 (1990), 551–558. doi:10.3758/BF03211601

    work page doi:10.3758/bf03211601 1990

  13. [13]

    Gescheider

    George A. Gescheider. 1997.Psychophysics: The Fundamentals. Lawrence Erlbaum Associates

    1997

  14. [14]

    Advances in Psychology, vol

    Sandra G. Hart and Lowell E. Staveland. 1988. Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research. InHuman Mental Workload, Peter A. Hancock and Najmedin Meshkati (Eds.). Advances in Psychology, Vol. 52. North-Holland, 139–183. doi:10.1016/S0166-4115(08)62386-9

    work page doi:10.1016/s0166-4115(08)62386-9 1988

  15. [15]

    Eve Hoggan, Stephen Brewster, and Jody Johnston. 2008. Investigating the Effectiveness of Tactile Feedback for Mobile Touchscreens. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’08). Association for Computing Machinery, 1573–1582. doi:10.1145/1357054.1357300

    work page doi:10.1145/1357054.1357300 2008

  16. [16]

    Neil Houlsby, Ferenc Huszár, Zoubin Ghahramani, and Máté Lengyel

  17. [17]

    Bayesian active learning for classification and preferenc e learning,

    Bayesian Active Learning for Classification and Preference Learning. arXiv:1112.5745 [stat.ML] https://arxiv.org/abs/1112.5745

    work page arXiv

  18. [18]

    Frederick A. A. Kingdom and Nicolaas Prins. 2016.Psychophysics: A Practical Introduction. Academic Press

    2016

  19. [19]

    Mohammad

    Svetlana Kiritchenko and Saif M. Mohammad. 2017. Best-Worst Scaling More Reliable than Rating Scales: A Case Study on Sentiment Intensity Annotation. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics (EACL): Volume 2, Short Papers. https://aclanthology. org/E17-2074/

    2017

  20. [20]

    Rensis Likert. 1932. A technique for the measurement of attitudes.Archives of psychology(1932)

    1932

  21. [21]

    Michael R Lin, Elena Makarova, Eytan Bakshy, and Peter I Frazier. 2022. Prefer- ence Exploration for Efficient Bayesian Optimization with Multiple Outcomes. InProceedings of the 25th International Conference on Artificial Intelligence and Statistics (AISTATS) (Proceedings of Machine Learning Research, Vol. 151)

    2022

  22. [22]

    Fontaine, Stefanos Nikolaidis, and Heather Culbertson

    Shihan Lu, Mianlun Zheng, Matthew C. Fontaine, Stefanos Nikolaidis, and Heather Culbertson. 2022. Preference-Driven Texture Modeling Through Interac- tive Generation and Search.IEEE Transactions on Haptics15, 3 (2022), 508–520

    2022

  23. [23]

    David J. C. MacKay. 1992. Information-Based Objective Functions for Active Data Selection.Neural Computation4, 4 (1992), 590–604. doi:10.1162/neco.1992.4.4.590

    work page doi:10.1162/neco.1992.4.4.590 1992

  24. [24]

    Anchit Mishra and Oliver Schneider. 2025. TacTalk: Personalizing Haptics Through Conversation. InProceedings of the 7th ACM Conference on Con- versational User Interfaces (CUI ’25). Association for Computing Machinery. doi:10.1145/3719160.3736638

    work page doi:10.1145/3719160.3736638 2025

  25. [25]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al

  26. [26]

    Training language models to follow instructions with human feedback

    Training Language Models to Follow Instructions with Human Feedback. arXiv:2203.02155

    work page internal anchor Pith review arXiv

  27. [27]

    Claudio Pacchierotti, Stephen Sinclair, Massimiliano Solazzi, Antonio Frisoli, Vincent Hayward, and Domenico Prattichizzo. 2017. Wearable Haptic Systems for the Fingertip and the Hand: Taxonomy, Review, and Perspectives.IEEE Transactions on Haptics10, 4 (2017), 580–600. doi:10.1109/TOH.2017.2689006

    work page doi:10.1109/toh.2017.2689006 2017

  28. [28]

    Shaoting Peng, Haonan Chen, and Katherine Driggs-Campbell. 2025. Towards Uncertainty Unification: A Case Study for Preference Learning. arXiv preprint arXiv:2503.19317

    work page arXiv 2025

  29. [29]

    Carl Edward Rasmussen and Christopher K. I. Williams. 2006.Gaussian Processes for Machine Learning. The MIT Press, Cambridge, MA, USA

    2006

  30. [30]

    Hasti Seifi, Matthew Chun, and Karon E. MacLean. 2018. Toward Affective Handles for Tuning Vibrations.ACM Transactions on Applied Perception (TAP) 15, 3 (2018), 1–23

    2018

  31. [31]

    Christiano

    Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, and Paul F. Christiano. 2020. Learning to Summarize from Human Feedback. InAdvances in Neural Information Processing Systems

    2020

  32. [32]

    L. L. Thurstone. 1927. A Law of Comparative Judgment.Psychological Review34, 4 (1927), 273–286

    1927

  33. [33]

    Christian Wirth, Riad Akrour, Gerhard Neumann, and Johannes Fürnkranz. 2017. A Survey of Preference-Based Reinforcement Learning Methods.Journal of Machine Learning Research18, 136 (2017), 1–46

    2017