Embed Size (px)
Citation preview
Nico [email protected] of MuensterGermanygo. wwu.de/buschlab
@nibroers
Bylinskii, Z,, Isola, P., Bainbridge, C, Torralba, & A., Oliva, A. (2015). Intrinsic and Extrinsic Effects on Image Memorability
Koen, J. D., Barrett, F. S., Harlow, I. M., & Yonelinas, A. P. (2016). The ROC Toolbox: A toolbox for analyzing receiver-operating characteristics derived from confidence ratings
Isola, P., Xiao, J., Torralba, A., Oliva, A. (2014), What makes an Image memorable?
Yonelinas, A .P. & Parks, C.M. (2007). Receiver Operating Characteristics (ROCs) in RecognitionMemory: A Review
Mem
orab
ility
Hig
hLo
w
Images were extracted from the data base established by Bylinskii et al (2015). We counterbalanced the set for semantic content (e.g., humans, animals) and employed the following memorability boundaries, whichwere more convservative than in previous research: highly memorable: >75%, medium memorable: 55-75%, low memorable: <55% hit rate.
Hig
h M
emo
Low Memo0.1 0.2 0.3 0.4
0.1
0.2
0.3
0.4
t(43)=-6.56; p < .001
-0.5 0 0.5 1 1.5-0.5
0
0.5
1
t(43)=-5.36; p < .001
Low Memo
Hig
h M
emo
Recollection. DPSD-Model recollection parameter is strongly associated with high memorability. Eachpoint represents one participant (n = 44)
Familiarity. Larger Familiarity for highly memorable pictures in DPSD-Model. Each point represents one participant (n=44).
The intrinsic memorability of pictures is associated with bothrecollection and familiarity Nico Broers & Niko Busch
Conclusion
0 0.5 1
Hit
Rate
0
0.5
1
strong encoding
FA Rate
weak encoding
Predicted ROCs. Concave in probability space and more asymmetrical (“Hockey-Stick“shape) along the chance diagonal for strong items
Observed ROCs. Highly memorable itemsare more asymmetrical along the chance diagonal, indicating better recollection for highly memorable images
0 0.5 1
Hit
Rate
0
0.5
1
high memolow memo
FA Rate
0 0.5 10
0.2
0.4
0.6
0.8
1
Area
und
er th
e cu
rve
r = .39; p < .001
Hit Rate
Mem
orab
ility
Area under the curve. Area under the curve increases linearly with memorability. Each point representsone image (n = 300).
Background
}
Institute of Psychology, University of Münster, Germany
Recent research has shown that many images are con-sistently remembered or forgotten by most people (Isola et al, 2014).
Any theory about the memorability of information shouldtake into account the phenomenology of remembering:it has long been acknowledged that old items can be recognized based on a feeling of familiarity or recollectionof specific details of the studied item.
This notion is supported by dual process signal detection (DPSD) models, which assume separate contributions of familiarity and recollection to receiver operating charac-teristics (ROCs) (e.g. Yonelinas, & Parks, 2007). Is the intrinsic memorability of an image associated witha general surge in familiarity or also an increase in recollection of the studied item?
ROCs were modeled with a dual process signal detection model (Koen et al, 2016). Note that the recollection and fa-miliarity parameters of this model were used as proxies for phenomenologically distinct ways of remembering; we remain neutral as to whether they represent distinct cogni-tive or neural mechanisms.
Area under the curve increased linearly with memorability. More importantly, ROC shapes indicated better recollection for highly memorable compared to non-memorable images. This effect was even stronger than the increase in familiarity for high memorability.
These findings suggest that recognition of memorable images is not only based on a feeling of having seen the image before, but on recollection of specific image details.
Time
+
Test
+ Old/New? Confidence1-6 +
Time500 ms 500 msSelf-Paced Self-Paced
Old/New? ...
...Encoding
+ Indoor/Outdoor? + Indoor/
Outdoor?
500 ms 500 msSelf-Paced Self-PacedVigilance TaskVigilance Task
Recollection
Familiarity
