A Model Relating Pupil Diameter to Mental Workload and Lighting Conditions
Open accessThe DOI matches its Crossref record; 49 of the 49 references it cites were found in a registry or an index.
Abstract
In this paper, we present a proof-of-concept approach to estimating mental workload by measuring the user's pupil diameter under various controlled lighting conditions. Knowing the user's mental workload is desirable for many application scenarios, ranging from driving a car, to adaptive workplace setups. Typically, physiological sensors allow inferring mental workload, but these sensors might be rather uncomfortable to wear. Measuring pupil diameter through remote eye-tracking instead is an unobtrusive method. However, a practical eye-tracking-based system must also account for pupil changes due to variable lighting conditions. Based on the results of a study with tasks of varying mental demand and six different lighting conditions, we built a simple model that is able to infer the workload independently of the lighting condition in 75% of the tested conditions.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“While fNIRS is famous for brain sensing in HCI research due to its easy setup [177], latencies exist between stimulus onset and the exact measure.”
From A Survey on Measuring Cognitive Workload in Human-Computer Interaction (2023) · cites this paper as [177]
Citations
Cites
Paced Auditory Serial-Addition Task: A Measure of Recovery from Concussion
Gronwall, D. M. A. (1977) · Perceptual and Motor Skills · doi:10.2466/pms.1977.44.2.367 · Cited by 2,395 · cited 1×
“Tasks used to induce mental workload, such as the N-back test, delayed digit recall test, or the PASAT [12, 17, 27] can be presented as either a visual stimulus or as an auditory stimulus.”
MethodDetecting Stress During Real-World Driving Tasks Using Physiological Sensors
Healey, J. A., & Picard, R. W. (2005) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2005.848368 · Cited by 2,207 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodTask-evoked pupillary responses, processing load, and the structure of processing resources.
Beatty, J. (1982) · Psychological Bulletin · doi:10.1037/0033-2909.91.2.276 · Cited by 1,959 · cited 3×
“A dated but thorough review of the literature surrounding task-evoked pupillary responses outlined the relationship between changes in pupil size and mental workload [3].”
BackgroundMethodAge differences in short-term retention of rapidly changing information.
Kirchner, W. K. (1958) · Journal of Experimental Psychology · doi:10.1037/h0043688 · Cited by 1,798 · cited 1×
“Tasks used to induce mental workload, such as the N-back test, delayed digit recall test, or the PASAT [12, 17, 27] can be presented as either a visual stimulus or as an auditory stimulus.”
MethodPupil Diameter and Load on Memory
Kahneman, D., & Beatty, J. (1966) · Science · doi:10.1126/science.154.3756.1583 · Cited by 1,617 · cited 1×
“A number of early studies supported this relationship between pupillary changes and workload, and in a way established the foundation for this approach to psychophysiology [15, 16].”
ResultAn introduction to modern missing data analyses
Baraldi, A. N., & Enders, C. K. (2010) · Journal of School Psychology · doi:10.1016/j.jsp.2009.10.001 · Cited by 1,452 · cited 1×
“Finally, we used a single imputation method [2] to fill in the remaining 3 averages: in our method the missing average is set equal to an average calculated for the same participant, and same lighting conditions, but a different task difficulty.”
MethodA breadth-first survey of eye-tracking applications
Duchowski, A. T. (2002) · Behavior Research Methods, Instruments, & Computers · doi:10.3758/bf03195475 · Cited by 1,061 · cited 1×
“Presentation Modalities of Workload Tasks In many studies where eye-tracking data were used as indices of mental workload, the task used to induce workload was often a visually-based stimulus [7].”
MethodFrontal midline EEG dynamics during working memory
Onton, J., Delorme, A., & Makeig, S. (2005) · NeuroImage · doi:10.1016/j.neuroimage.2005.04.014 · Cited by 850 · cited 6×
“Frontal midline EEG dynamics during working memory.”
MethodInfluentialNeurophysiological measures of cognitive workload during human-computer interaction
Gevins, A., & Smith, M. E. (2003) · Theoretical Issues in Ergonomics Science · doi:10.1080/14639220210159717 · Cited by 414
Estimating workload using EEG spectral power and ERPs in the n-back task
Brouwer, A.-M., Hogervorst, M. A., van Erp, J. B. F., Heffelaar, T., Zimmerman, P. H., & Oostenveld, R. (2012) · Journal of Neural Engineering · doi:10.1088/1741-2560/9/4/045008 · Cited by 407 · cited 6×
“Frontal midline EEG dynamics during working memory.”
MethodInfluentialRyu, K., & Myung, R. (2005) · International Journal of Industrial Ergonomics · doi:10.1016/j.ergon.2005.04.005 · Cited by 399 · cited 6×
“Frontal midline EEG dynamics during working memory.”
MethodInfluentialMehler, B., Reimer, B., Coughlin, J. F., & Dusek, J. A. (2009) · Transportation Research Record: Journal of the Transportation Research Board · doi:10.3141/2138-02 · Cited by 392
Estimating cognitive load using remote eye tracking in a driving simulator
Palinko, O., Kun, A. L., Shyrokov, A., & Heeman, P. (2010) · Proceedings of the 2010 Symposium on Eye-Tracking Research & Applications - ETRA '10 · doi:10.1145/1743666.1743701 · Cited by 371 · cited 1×
“However, Marshall’s proprietary algorithm by default outputs a sequence of estimates once a second, which can obscure interesting changes in cognitive activity that occur at a quicker pace, such as those observed in spoken interactions [14, 22, 36].”
BackgroundSelf-citationUsing eye movement activity as a correlate of cognitive workload
Ahlstrom, U., & Friedman-Berg, F. J. (2006) · International Journal of Industrial Ergonomics · doi:10.1016/j.ergon.2006.04.002 · Cited by 309 · cited 2×
“Ahlstrom & Friedman-Berg [1] found that air traffic controllers exhibited greater pupil diameter and shorter blink duration in response to increased task difficulty.”
BackgroundDriver workload and eye blink duration
Benedetto, S., Pedrotti, M., Minin, L., Baccino, T., Re, A., & Montanari, R. (2011) · Transportation Research Part F: Traffic Psychology and Behaviour · doi:10.1016/j.trf.2010.12.001 · Cited by 303 · cited 1×
“Benedetto and colleagues [4] also recently found that increased dual-task difficulty was accompanied by shorter blink duration and longer duration between blinks in a simulated driving environment, though the former provided more reliable data.”
BackgroundMehler, B., Reimer, B., & Coughlin, J. F. (2012) · Human Factors: The Journal of the Human Factors and Ergonomics Society · doi:10.1177/0018720812442086 · Cited by 294 · cited 1×
“Mehler and colleagues found that heart rate and skin conductance were reliably indicative of driver workload estimates in simulated driving tasks [25, 26].”
BackgroundEye Activity Correlates of Workload during a Visuospatial Memory Task
Van Orden, K. F., Limbert, W., Makeig, S., & Jung, T.-P. (2001) · Human Factors: The Journal of the Human Factors and Ergonomics Society · doi:10.1518/001872001775992570 · Cited by 293
The Index of Cognitive Activity: measuring cognitive workload
Marshall, S. P. (2003) · Proceedings of the IEEE 7th Conference on Human Factors and Power Plants · doi:10.1109/hfpp.2002.1042860 · Cited by 266
Klingner, J., Tversky, B., & Hanrahan, P. (2011) · Psychophysiology · doi:10.1111/j.1469-8986.2010.01069.x · Cited by 255 · cited 1×
“[19] found that the difficulty of a task may change between different perceptual modalities.”
BackgroundPupil location under mesopic, photopic, and pharmacologically dilated conditions.
Yang, Y., Thompson, K. P., & Burns, S. A. (2002) · PubMed · Cited by 230
Measuring the task-evoked pupillary response with a remote eye tracker
Klingner, J., Kumar, R., & Hanrahan, P. (2008) · Proceedings of the 2008 symposium on Eye tracking research & applications - ETRA '08 · doi:10.1145/1344471.1344489 · Cited by 222 · cited 1×
“One potential solution to this issue could be to introduce an auditory-based workload task and have participants focus on a single point in a visually neutral or uninteresting stimulus [18].”
BackgroundTask performance and eye activity: predicting behavior relating to cognitive workload.
Tsai, Y.-F. D., Viirre, E., Strychacz, C., Chase, B., & Jung, T. (2007) · PubMed · Cited by 218
Assessing pilot workload. Why measure heart rate, HRV and respiration?
Roscoe, A. H. (1992) · Biological Psychology · doi:10.1016/0301-0511(92)90018-p · Cited by 213 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodEye movements and blinks: their relationship to higher cognitive processes
Fogarty, C., & Stern, J. A. (1989) · International Journal of Psychophysiology · doi:10.1016/0167-8760(89)90017-2 · Cited by 201 · cited 1×
“Other early research with eye-tracking indicated a greater time interval between blinks for a visual identification task compared to a visual detection task [8].”
BackgroundHeart rate variability: indicator of user state as an aid to human-computer interaction
Rowe, D. W., Sibert, J., & Irwin, D. (1998) · Proceedings of the SIGCHI conference on Human factors in computing systems - CHI '98 · doi:10.1145/274644.274709 · Cited by 192 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodShifting Gears: User Interfaces in the Age of Autonomous Driving
Kun, A. L., Boll, S., & Schmidt, A. (2016) · IEEE Pervasive Computing · doi:10.1109/mprv.2016.14 · Cited by 158
Self-citationCognitive demands of hands-free-phone conversation while driving
Nunes, L., & Recarte, M. A. (2002) · Transportation Research Part F Traffic Psychology and Behaviour · doi:10.1016/s1369-8478(02)00012-8 · Cited by 147 · cited 1×
“Nunes & Recarte [32] found increased pupil diameter and higher blink rates when participants were engaged in more difficult secondary tasks while driving an automobile.”
BackgroundReal-time eye detection and tracking under various light conditions
Zhu, Z., Fujimura, K., & Ji, Q. (2002) · ACM Symposium on Eye Tracking Research and Applications (ETRA) · doi:10.1145/507072.507100 · Cited by 146 · cited 1×
“Methods have also been developed to gather eye gaze data in sub-optimal lighting conditions [49, 50].”
A data set of real world driving to assess driver workload
Schneegass, S., Pfleging, B., Broy, N., Heinrich, F., & Schmidt, A. (2013) · Proceedings of the 5th International Conference on Automotive User Interfaces and Interactive Vehicular Applications · doi:10.1145/2516540.2516561 · Cited by 135 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodSelf-citationBrainput: enhancing interactive systems with streaming fnirs brain input
Solovey, E., Schermerhorn, P., Scheutz, M., Sassaroli, A., Fantini, S., & Jacob, R. (2012) · Proceedings of the SIGCHI Conference on Human Factors in Computing Systems · doi:10.1145/2207676.2208372 · Cited by 86 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodReimer, B., Mehler, B., Coughlin, J. F., Godfrey, K. M., & Tan, C. (2009) · Proceedings of the 1st International Conference on Automotive User Interfaces and Interactive Vehicular Applications · doi:10.1145/1620509.1620531 · Cited by 75
A comparison of two scoring procedures with the NASA task load index in a simulated flight task
Moroney, W. F., Biers, D. W., Eggemeier, F. T., & Mitchell, J. A. (2003) · Proceedings of the IEEE 1992 National Aerospace and Electronics Conference@m_NAECON 1992 · doi:10.1109/naecon.1992.220513 · Cited by 62 · cited 1×
“Unweighted subjective workload scores for each participant were calculated by averaging participants’ ratings on a scale of 1-20 over the six dimensions of the NASA-TLX questionnaire [29].”
MethodPalinko, O., & Kun, A. L. (2012) · Proceedings of the Symposium on Eye Tracking Research and Applications · doi:10.1145/2168556.2168650 · Cited by 59 · cited 1×
“A series of recent preliminary studies have addressed this issue in further detail, in the context of a simulated driving environment [23, 34, 35].”
BackgroundSelf-citationMyrtek, M., Weber, D., Brügner, G., & Müller, W. (1996) · Biological Psychology · doi:10.1016/0301-0511(95)05168-6 · Cited by 51 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodEffects of Grouping on the Pupillary Response in a Short-Term Memory Task
Kahneman, D., Onuska, L., & Wolman, R. E. (1968) · Quarterly Journal of Experimental Psychology · doi:10.1080/14640746808400168 · Cited by 51
Exploring the Influence of Light and Cognitive Load on Pupil Diameter in Driving Simulator Studies
Palinko, O., & Kun, A. L. (2011) · International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design (Driving Assessment Conference) · doi:10.17077/drivingassessment.1416 · Cited by 48
Self-citationGirouard, A., Solovey, E. T., & Jacob, R. J. K. (2013) · International Journal of Autonomous and Adaptive Communications Systems · doi:10.1504/ijaacs.2013.050689 · Cited by 39 · cited 1×
“Other Physiological Measures of Mental Workload Physiological measures such as functional near-infrared spectroscopy (fNIRS) [11, 44], electroencephalography EEG [5, 10, 33, 42], heart rate [31, 40] and heart rate variability [31, 41], and skin conductance [13, 43], have also been used to track mental workload while e…”
MethodGable, T. M., Kun, A. L., Walker, B. N., & Winton, R. J. (2015) · International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI) · doi:10.1145/2809730.2809745 · Cited by 39 · cited 1×
“found evidence that in driving simulator studies pupil diameter might be a more sensitive measure of cognitive load changes than heart rate [9].”
BackgroundSelf-citationKun, A. L., Palinko, O., Medenica, Ž., & Heeman, P. A. (2013) · Interspeech (USB) · doi:10.21437/interspeech.2013-595 · Cited by 39 · cited 2×
“After all, prior research shows that after the end of a cognitively-demanding task, pupil diameter returns to its pre-task level within a few seconds (see for instance [3, 22]).”
BackgroundMethodSelf-citationExploring the effects of size and luminance of visual targets on the pupillary light reflex
Kun, A. L., Palinko, O., & Razumenić, I. (2012) · International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI) · doi:10.1145/2390256.2390287 · Cited by 30
Self-citationDriver wellness, safety and the development of an AwareCar
Coughlin, J. F., Reimer, B., & Mehler, B. L. (2009) · Cited by 28
Estimating cognitive load using pupil diameter during a spoken dialogue task
Heeman, P. A., Meshorer, T., Kun, A. L., Palinko, O., & Medenica, Ž. (2013) · International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI) · doi:10.1145/2516540.2516570 · Cited by 23 · cited 1×
“However, Marshall’s proprietary algorithm by default outputs a sequence of estimates once a second, which can obscure interesting changes in cognitive activity that occur at a quicker pace, such as those observed in spoken interactions [14, 22, 36].”
BackgroundSelf-citationExperiencing Autonomous Vehicles
Meschtscherjakov, A., Tscheligi, M., Szostak, D., Ratan, R., McCall, R., Politis, I., & Krome, S. (2015) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/2702613.2702661 · Cited by 23
Exploring user expectations for context and road video sharing while calling and driving
Pfleging, B., Schneegaß, S., & Schmidt, A. (2013) · International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI) · doi:10.1145/2516540.2516547 · Cited by 14 · cited 1×
“One example is the automotive domain where it is important not to overload the driver as they interact with in-vehicle devices [28] or communicate with remote conversants [21, 38], since performance degrades with increasing workload [6].”
BackgroundSelf-citationYang, Y., Reimer, B., Mehler, B. L., & Dobres, J. (2013) · International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design (Driving Assessment Conference) · doi:10.17077/drivingassessment.1518 · Cited by 11 · cited 1×
“In an on-road driving study they also found similar changes in heart rate and eye gaze measurements with respect to task difficulty [48].”
ResultToward Usable Intelligent User Interface
Mezhoudi, N., Khaddam, I., & Vanderdonckt, J. M. (2015) · Lecture notes in computer science · doi:10.1007/978-3-319-20916-6_43 · Cited by 11
Miller, W. T., & Kun, A. L. (2013) · International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI) · doi:10.1145/2516540.2516564 · Cited by 10 · cited 1×
“One example is the automotive domain where it is important not to overload the driver as they interact with in-vehicle devices [28] or communicate with remote conversants [21, 38], since performance degrades with increasing workload [6].”
BackgroundSelf-citationVideo call, or not, that is the question
Kun, A. L., & Medenica, Ž. (2012) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/2212776.2223684 · Cited by 8
Self-citationThe impact of repeated cognitive tasks on driving performance and visual attention
Wang, Y., Reimer, B., Mehler, B. L., Zhang, J., Mehler, A. C., & Coughlin, J. F. (2010) · Advances in human factors and ergonomics series · doi:10.1201/ebk1439834992-70 · Cited by 8 · cited 1×
“The same research group showed that pupil diameter can be used to estimate workload for drivers engaged in the N-back task [47].”
Background
Research works printed but not linked: 6
- 10.1037/0033--2909.91.2.276 · doi:10.1037/0033--2909.91.2.276
- 10.1088/1741--2560/9/4/045008 · doi:10.1088/1741--2560/9/4/045008
- 10.1016/0167--8760(89)90017--2 · doi:10.1016/0167--8760(89)90017--2
- 10.1111/j.1469--8986.2010.01069.x · doi:10.1111/j.1469--8986.2010.01069.x
- 10.1016/0301-0511(95)05168--6 · doi:10.1016/0301-0511(95)05168--6
- 10.1016/s1369--8478(02)00012--8 · doi:10.1016/s1369--8478(02)00012--8
Cited by
A Review of Psychophysiological Measures to Assess Cognitive States in Real-World Driving
Lohani, M., Payne, B. R., & Strayer, D. L. (2019) · Frontiers in Human Neuroscience · doi:10.3389/fnhum.2019.00057 · Cited by 348
“to infer cognitive workload after accounting for some variations in lighting conditions (Pfleging et al., 2016; Reilly et al., 2018).”
BackgroundA Survey on Measuring Cognitive Workload in Human-Computer Interaction
Kosch, T., Karolus, J., Zagermann, J., Reiterer, H., Schmidt, A., & Woźniak, P. W. (2023) · ACM Computing Surveys · doi:10.1145/3582272 · Cited by 304
“Thus, different methods and models (e.g., [195]) were developed to ensure a valid measurement of cognitive workload using pupil dilation: Here, the total pupil dilation is seen as the sum of dilation due to light intensity and cognitive workload.”
Self-citationContext-Aware Online Adaptation of Mixed Reality Interfaces
Lindlbauer, D., Feit, A. M., & Hilliges, O. (2019) · Proceedings of the 32nd Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3332165.3347945 · Cited by 254
“Since pupil dilation is confounded with ambient light (cf. [41]), several metrics exist that do not measure absolute size but change in size, such as the proprietary Index of Cognitive Activity et al. [31, 5], and the Index of Pupillary Activity (IPA) by Duchowski et al. [10], which we use in our…”
MethodDuchowski, A. T., Krejtz, K., Krejtz, I., Biele, C., Niedzielska, A., Kiefer, P., Raubal, M., & Giannopoulos, I. (2018) · Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3173574.3173856 · Cited by 228
“[53] and Palinko and Kun [50] provide notable examples related to HCI, including automotive and online learning domains.”
BackgroundMeasuring Cognitive Load using Eye Tracking Technology in Visual Computing
Zagermann, J., Pfeil, U., & Reiterer, H. (2016) · Proceedings of the Sixth Workshop on Beyond Time and Errors on Novel Evaluation Methods for Visualization · doi:10.1145/2993901.2993908 · Cited by 224
Metadata
- Authors
- Pfleging, B., Fekety, D. K., Schmidt, A., & Kun, A. L.
- Affiliations
- University of Stuttgart (DE)
- Clemson University (US)
- University of New Hampshire (US)
- Venue
- Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems
- Publisher
- ACM
- Published
- 2016-05-07
- Pages
- 5776-5788
- OpenAlex
- W2346312659
- Cited by
- 179
- References
- 49
- Citations by year
- 2026: 11 · 2025: 22 · 2024: 20 · 2023: 18 · 2022: 20 · 2021: 24 · 2020: 20 · 2019: 11 · 2018: 21 · 2017: 6 · 2016: 6
- FWCI
- 10.01
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- CHI Conference on Human Factors in Computing Systems (publishedVersion)
- Funders
- Deutsche Forschungsgemeinschaft (EXC 310/2 and SFB/Transregio 161, project C02)
- National Science Foundation (IIA-1358096)
- Topics
- Physical Sciences › Computer Science › Human-Computer Interaction › Gaze Tracking and Assistive Technology
- Social Sciences › Human Factors and Ergonomics › Human-Automation Interaction and Safety
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Keywords
- mental workload estimation, pupil size, remote eye tracking, lighting conditions, workload modeling, driving
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.