Representation of multiple objects
- Bao, P., & Tsao, D. Y. (2018). Representation of multiple objects in macaque category-selective areas. Nature Communications, article
- How our visual experience is clutter invariant?
- Neurons in the Macaque IT complex showed winner-take-all normalization (similar to others like Zoccolan2015's weighted averaging); Middle body patch and face patch (middle lateral face patch). They included category and spatial selectivity of neighboring neurons parameter to the normalization model. Weights are defined by neuron's object preference (w1 = 1, w2 = 0, if 1 is preferred).
- Face (L) + Object (R) -> measured at the right face area (which means contralaterally processed) = leads to winner-take-all, non-face object at the ipsilateral didn't affect the activity. But, when Object (L) + Face (R), so face is at ipsilateral and non-face at contra = leads to averaging. And same organization, if vertically presented (within hemifield), leads to winner-take-all. Face cells followed a winner-take-all for both upper & lower when presented with non-face.
- Ruled out the role of spatial tuning by ruling out at cells that are selectively responsive to ipsi, and they found the same results.
- In addition to faces, they showed the same for body images.
- Similar to Busse2009's findings of how increase in relative contrast lead to winner-take-all (from averaging) at V1, this study showed the same for face presented at ipsi.
- Rolls, EdmundT., & Tovee, MartinJ. (1995). The responses of single neurons in the temporal visual cortical areas of the macaque when more than one stimulus is present in the receptive field. Experimental Brain Research, article
- They mention about "translational invariance effect", where a group of neurons' responses does not change as long as the stimuli hit the RF (no matter where it occupies).
- Then they asked if other stimuli/objects that fall outside of the RF change neurons' activity in response to that RF. Monkey brains and spiking rates.
- Results show: interaction, even averaging, and responses got scaled and non-linear (non-additive).
- For example when attention is deployed to an "effective" stimuli at parafovea, if there is an ineffective stimulus at the unattended region, responses drop and we see an averaging of the activity (they are bound).
- Bles, M., Schwarzbach, J., De Weerd, P., Goebel, R., & Jansma, B. M. (2006). Receptive field size-dependent attention effects in simultaneously presented stimulus displays. NeuroImage, article
- When multiple objects are displayed, their proximity to each other determines the level of interaction they have. And this is linked with the object encoding receptive field (RF) sizes. For example, if objects are dispersed a little, V1 neurons would not respond to that dispersion as their RF is relatively smaller.
- Attend vs unattend, and measured with fMRI at early ve high-level visual areas (V4, TEO). Stimuli are separated by 2, 4, and 7 degrees.
- Result: They observe a biased competition model. Attended - unattended leads to greater differences with increasing processing level (V4 can encode 4 degrees of RF). If stimuli fall into their RF -> biased competition.
∅ Authors' illustration of their hypothesis: objects proximity, biased competition and receptive field sizes.
- Won, B.-Y., & Jiang, Y. V. (2013). Redundancy effects in the processing of emotional faces. Vision Research, article
- Seeing more (both same or different) is better than seeing it alone (in terms of detecting the facial expression) - d prime. And RT goes down also.
- If faces are Fearful, seeing 4DIFF > 4SAME
- Neutral or fearful (emotion detection task) in the first experiment. 4-SAME and 4-DIFF equally improved the performance.
- Experiment 2: male or female task, facial expressions are the same, fearful and neutral differently bucketed.
- Results: seeing "more" (4-SAME & 4-DIFF) leads to higher discrimination sensitivity - redundancy gain.
- Experiment 3: is redundancy gain result of statistical summation or perceptual summation. Seeing more (4 vs 1) did not lead to higher face-after-effects. Therefore, statistical summation rules over the perceptual summation - therefore redundancy gain might not be rooted from elavated neural activity in response to seeing 4 faces, because if it was, one would see a larger FAE but they did not observe that..
- Joo, S. J., Boynton, G. M., & Murray, S. O. (2012). Long-Range, Pattern-Dependent Contextual Effects in Early Human Visual Cortex. Current Biology, article
- Neurons encoding the center (target) are affected by the high-level, global features of an image.
- When context (flankers) deviate from the center, responses to center is enhanced.
- Responses are highest when the target image flanks with orthogonal oriented gratings in contrast detection task using sensory adaptation paradigm (more activity = increased threshold after adaptation).
- Alternating pattern like, HVHVH < orthogonal pattern.
- VHV > VVV/HHH
- Electrophysiology (EEG), fMRI, psychophysics, all together, pretty neat study!
- Jiang, Y. V., Kwon, M., Shim, W. M., & Won, B.-Y. (2010). Redundancy effects in the perception and memory of visual objects. Visual Cognition, article
- Memory (RT and accuracy) on representing multiple faces.
- If faces are identical, 4SAME > SINGLE (classical ensemble perception finding).
- Multiple identical items lead to enhanced perceptual priming and visual short-term and long-term memory, a redundancy effect. 4-DIFF did not always lead to enhanced performance.
- Interesting finding here, gender of the faces seems to be encoded independently: If prime and test have incongruency (in terms of gender), then priming with 4-DIFFERENT faces increases RT compared to seeing 4-SAME and SINGLE.
- Made me think about differences in conceptually vs perceptually processing.
- Cohen, M. A., Rhee, J. Y., & Alvarez, G. A. (2016). Limits on perceptual encoding can be predicted from known receptive field properties of human visual cortex. Journal of Experimental Psychology: Human Perception and Performance, article
- Spatial interference effect, similar to Bles2006, RF sizes of areas (and groups of neurons in it of course) can provide long-range interference.
- IT receptive fields have strong contralateral bias
- Receptive fields beyond V4 do often cross the vertical meridian (Golomb & Kanwisher, 2012; Kravitz et al., 2008; Wan- dell et al., 2007)
- Shows ability to encode multiple items simultaneously is limited, leading to a drop in discrimination threshold (task: change detection).
- Experiment 1: within vs across hemifield with objects, scene, faces (all high-level), and colors and orientation, found out that the performance for the across-hemi is better.
- Experiment 2: within hemifield but far vs close.
- Experiment 3: same as 2 but colors and objects, more closer.
- Experiment 4: simultaneous vs sequential presentation of stimuli, if spatial interference due to RF is the thing here, it should only work for simultaneous presentation, and that is what they found.
- Looks like a good behavioral follow-up study of Zoccolan 2015 monkey paper.
- Afraz, S.-R., & Cavanagh, P. (2008). Retinotopy of the face aftereffect. Vision Research, article
- Study of aftereffect zone. Face recognition compromised with dual stimulus presentation.
- Significant negative correlation between the face-after-effect (FAE) strength with decreases in the interstimulus distance.
- When the distance between stimuli increases they saw a larger FAE (meaning more competition, similar to "biased competition" - Desimone 1998).
- Test of simultaneous vs sequential presentation:
- FAE is strong at the sim condition.
- Take home: FAE significantly drops when faces are closer to each other (within 10 degrees and 3 eccentricities more stim is leading to competition (SPE shifts rightwards). This is similar to the IT neural pattern.
- Puce, A., McNeely, M. E., Berrebi, M. E., Thompson, J. C., Hardee, J., & Brefczynski-Lewis, J. (2013). Multiple faces elicit augmented neural activity. Frontiers in Human Neuroscience, article
- Redundancy gain was observed in the N170 component. More faces -> larger response amplitude.
- Zoccolan, D. (2005). Multiple Object Response Normalization in Monkey Inferotemporal Cortex. Journal of Neuroscience, 25(36), article
- IT responses in clutter: are they clutter tolerant?
- Response normalization is at work for IT as well. Responses to object pairs were smaller than the simple sum of individual responses (A & B < A + B (in isolation)).
"Responses are normalized in that they were approximately the average of the constituent object's responses".
Responses do not change as a function of RF position.
Including another object interferes and drops down the overall response (authors also argue this would be reflected in recognition performance, very neat ideas!).
- Shim, W. M., Jiang, Y. V., & Kanwisher, N. (2013). Redundancy gains in retinotopic cortex. Journal of Neurophysiology, article
- fMRI study, seeing many faces lead to increased BOLD. And authors argue this is not about overlapping receptive fields but overlapping in higher-level category representation because 4-same > 4-diff (in BOLD)
- Sim vs sequ presentation: 4-same sequ led to increased activity in high-level regions while 4-same sim lead to increased activity in the early visual cortex. Very interesting!
Can be referred to as a neural basis to Sweeny2013 findings.
- Busse, L., Wade, A. R., & Carandini, M. (2009). Representation of Concurrent Stimuli by Population Activity in Visual Cortex. Neuron, 64(6), article
- Very interesting study! Cat braints :( in response to seeing two gratings measured at V1. As the relative contrast between them increases, neural encoding strategy shifts from averaging to winner-take-all. And/or there is weighted averaging system and weights become 1 if one of the grating has a robust contrast difference.
- Winner-take-all competition: where the response to the summed stimuli resembles the response to the stronger stimulus alone.
- Test of human V1 and MT using EEG provided complimentary findings.