the dimensionality of beauty 1 qihan wu ,aenne a 2 ... · the dimensionality of beauty.18th annual...

The dimensionality of beauty Qihan Wu 1 , Aenne A. Brielmann 1 , Mika S. Simoncelli 3 & Denis G. Pelli 1, 2 Is beauty one-dimensional? Acknowledgements References Contact Please contact [email protected] for more information. We thank Larry Maloney, Katerina Malakhova, Hortense Gimonet, Najib Majaj and Mika Simoncelli for helpful comments. It’s amazing that such a simple model fits so well. Beauty-difference ratings are well fit by a one-dimensional model. Thus, human mean relative-beauty ratings are consistent. This is an important consideration for the potential of beauty judgments to underlie decision making. Conclusion: A 1-D model fits beauty difference ratings well. [1] Kurdi, B., Lozano, S., & Banaji, M. R. (2016). Introducing the Open Affective Standardized Image Set (OASIS). Behavior research methods, 1-14. “Which is more beautiful, and by how much?” 14 images from OASIS [1] database, for which we have normative valence and beauty ratings. Stimuli 6 self-selected images (3 beautiful and 3 not beautiful). Procedure Our 1-D model Are human mean relative-beauty judgments consistent with representation of each object’s beauty by one number? That would mean that there is a linear beauty scale line, and each object gets one point on that line. If this is right, then participants’ mean relative- beauty judgments will never be contradictory. All the possible pairs among 20 images were presented to each participant twice for test-retest. Each participant did 380 trials. For each trial, the participant chose which image was more beautiful and indicated by how much on a scale from 1 to 9. We hypothesize that the observer samples an estimate of beauty from a normal distribution. For one observer looking at 20 images, our model has 21 free parameters, one mean beauty rating per image and one common variance ( 2 ). We found the values of the 21 parameters that maximize the likelihood of the beauty-difference ratings. We took half of the data to fit the model, and predicted the rest. The average correlation was r = 0.76 (dashed line is the equality line). Test-retest differences are small. Participants had consistent beautydifference ratings. Maximum likelihood estimation (MLE) fits well. Difference rating Participants are consistent. Image Image Citation http://psych.nyu.edu/pelli/posters.html Wu, Q., Brielmann, A. A., Simoncelli, M. S., & Pelli, D. G. The dimensionality of beauty.18th Annual Meeting of the Vision Sciences Society, St. Pete Beach, Florida, May 18-23, 2018. Beauty difference rating Predicted beauty difference rating Predicted beauty difference rating Beauty difference rating Count Count Beauty difference rating changes Beauty difference rating changes 3 Stuyvesant High School, New York, NY, 10282 1 Psychology & 2 Neural Science, New York University, New York, NY 10003

Upload: others

Post on 03-Jul-2020

0 views

Category:

Documents

0 download

Report

Download

Embed Size (px):

TRANSCRIPT

Page 1: The dimensionality of beauty 1 Qihan Wu ,Aenne A 2 ... · The dimensionality of beauty.18th Annual Meeting of the Vision Sciences Society, St. Pete Beach, Florida, May 1 8-23, 201

The dimensionality of beauty Qihan Wu1, Aenne A. Brielmann1, Mika S. Simoncelli3 & Denis G. Pelli1, 2

Is beauty one-dimensional?

Acknowledgements

References

ContactPlease contact [email protected] for moreinformation.

We thank Larry Maloney, KaterinaMalakhova, Hortense Gimonet, Najib Majajand Mika Simoncelli for helpful comments.

It’s amazing that sucha simple model fits sowell. Beauty-differenceratings are well fit by aone-dimensional model.

Thus, human mean relative-beauty ratingsare consistent. This is animportant considerationfor the potential ofbeauty judgments tounderlie decision making.

Conclusion: A 1-Dmodel fits beautydifference ratingswell.

[1] Kurdi, B., Lozano, S., & Banaji, M. R. (2016). Introducing the Open Affective Standardized Image Set (OASIS). Behavior research methods, 1-14.

“Which is more beautiful,and by how much?”

14 images from OASIS [1] database,for which we have normative valenceand beauty ratings.

Stimuli

6 self-selected images (3 beautifuland 3 not beautiful).

Procedure

Our 1-D modelAre human mean relative-beauty judgments

consistent with representation of each object’s beautyby one number? That would mean that there is a linearbeauty scale line, and each object gets one point onthat line. If this is right, then participants’ mean relative-beauty judgments will never be contradictory.

All the possible pairs among 20 images were presented toeach participant twice for test-retest. Each participant did 380trials. For each trial, the participant chose which image was morebeautiful and indicated by how much on a scale from 1 to 9.

We hypothesize that the observer samples an estimateof beauty from a normal distribution.

For one observer looking at 20 images, our model has21 free parameters, one mean beauty rating per imageand one common variance (𝜎2). We found the values ofthe 21 parameters that maximize the likelihood of thebeauty-difference ratings.

We took halfof the data to fitthe model, andpredicted the rest.The averagecorrelation wasr = 0.76 (dashedline is the equalityline).

Test-retestdifferencesare small.Participantshad consistentbeauty-differenceratings.

Maximum likelihood estimation (MLE) fits well.

Difference rating

Participants are consistent.

Image Image

Citation

http://psych.nyu.edu/pelli/posters.html

Wu, Q., Brielmann, A. A., Simoncelli, M. S., & Pelli, D. G. The dimensionality of beauty.18th Annual Meeting of the Vision Sciences Society, St. Pete Beach, Florida, May 18-23, 2018.

Beauty difference rating

Pred

icte

dbe

auty

diffe

renc

era

ting

Pred

icte

dbe