Bridging Neurons and Behavior in a Convolutional Neural Network with Emergent Human-like Covert Attention
ABSTRACT Covert visual attention allows the brain to select different regions of the visual world without eye movements. Cues predictive of a target location orient covert attention and improve perceptual performance. How entire neuronal populations represent and integrate target, cues, and location information to result in behavioral signatures of covert attention is not understood. We analyze 1.8M neurons of feedforward Convolutional Neural Networks (CNNs) that show human-like attentional cueing effects. Consistent with neurophysiology, we show early layers with retinotopic neurons separately tuned to target or cue, and later layers with neurons with joint tuning and increased influence of the cue on target responses. We show cue-inhibitory and location-opponent neurons that are unreported by neurophysiologists and identify the neuronal circuits that result in neuronal cueing effects. The cue influences the mean neuronal responses and changes target sensitivity through three mechanisms: cue-weighted summation and opponency across locations, and interaction with the thresholding Rectified Linear Unit (ReLU). Some CNN computational stages mirror a Bayesian ideal observer (BIO), but with more gradual transitions, while the opponency and ReLU interaction are distinct from the BIO. Together, the findings establish a likely system-wide characterization of the brain computations that mediate the behavioral signatures of covert attention..
| Medienart: |
|
|---|
Preprint| Erscheinungsjahr: |
2024 |
|---|---|
| Erschienen: |
2024 |
| Enthalten in: |
bioRxiv.org - (2024) vom: 01. Apr. Zur Gesamtaufnahme - year:2024 |
|---|
| Sprache: |
Englisch |
|---|
| Beteiligte Personen: |
Srivastava, Sudhanshu [VerfasserIn] |
|---|
| Links: |
Volltext [kostenfrei] |
|---|
| Themen: |
|---|
| doi: |
10.1101/2023.09.17.558171 |
|---|
| funding: |
|
|---|---|
| Förderinstitution / Projekttitel: |
|
| PPN (Katalog-ID): |
XBI040903745 |
|---|
| LEADER | 01000caa a22002652 4500 | ||
|---|---|---|---|
| 001 | XBI040903745 | ||
| 003 | DE-627 | ||
| 005 | 20240405120407.0 | ||
| 007 | cr uuu---uuuuu | ||
| 008 | 230920s2024 xx |||||o 00| ||eng c | ||
| 024 | 7 | |a 10.1101/2023.09.17.558171 |2 doi | |
| 035 | |a (DE-627)XBI040903745 | ||
| 035 | |a (biorXiv)10.1101/2023.09.17.558171 | ||
| 040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
| 041 | |a eng | ||
| 100 | 1 | |a Srivastava, Sudhanshu |e verfasserin |0 (orcid)0000-0002-4721-0641 |4 aut | |
| 245 | 1 | 0 | |a Bridging Neurons and Behavior in a Convolutional Neural Network with Emergent Human-like Covert Attention |
| 264 | 1 | |c 2024 | |
| 336 | |a Text |b txt |2 rdacontent | ||
| 337 | |a Computermedien |b c |2 rdamedia | ||
| 338 | |a Online-Ressource |b cr |2 rdacarrier | ||
| 520 | |a ABSTRACT Covert visual attention allows the brain to select different regions of the visual world without eye movements. Cues predictive of a target location orient covert attention and improve perceptual performance. How entire neuronal populations represent and integrate target, cues, and location information to result in behavioral signatures of covert attention is not understood. We analyze 1.8M neurons of feedforward Convolutional Neural Networks (CNNs) that show human-like attentional cueing effects. Consistent with neurophysiology, we show early layers with retinotopic neurons separately tuned to target or cue, and later layers with neurons with joint tuning and increased influence of the cue on target responses. We show cue-inhibitory and location-opponent neurons that are unreported by neurophysiologists and identify the neuronal circuits that result in neuronal cueing effects. The cue influences the mean neuronal responses and changes target sensitivity through three mechanisms: cue-weighted summation and opponency across locations, and interaction with the thresholding Rectified Linear Unit (ReLU). Some CNN computational stages mirror a Bayesian ideal observer (BIO), but with more gradual transitions, while the opponency and ReLU interaction are distinct from the BIO. Together, the findings establish a likely system-wide characterization of the brain computations that mediate the behavioral signatures of covert attention. | ||
| 650 | 4 | |a Biology |7 (dpeaa)DE-84 | |
| 650 | 4 | |a 570 |7 (dpeaa)DE-84 | |
| 700 | 1 | |a Wang, William Yang |e verfasserin |4 aut | |
| 700 | 1 | |a Eckstein, Miguel P. |e verfasserin |4 aut | |
| 773 | 0 | 8 | |i Enthalten in |t bioRxiv.org |g (2024) vom: 01. Apr. |
| 773 | 1 | 8 | |g year:2024 |g day:01 |g month:04 |
| 856 | 4 | 0 | |u http://dx.doi.org/10.1101/2023.09.17.558171 |m X:VERLAG |x 0 |z kostenfrei |3 Volltext |
| 912 | |a GBV_XBI | ||
| 951 | |a AR | ||
| 952 | |j 2024 |b 01 |c 04 | ||