Stochastic and synchronous neural circuit dynamics underlying feature-similarity gain modulation

Stochastic and synchronous neural circuit dynamicsunderlying feature-similarity gain modulation

Xiao-Jing Wang With Salvador Ardid and Albert Compte

Compte, Brunel, Goldman-Rakic and Wang 2000

A recurrent network mechanism for working memory

A network model of PFC/PPC-MT loop

Ardid et al J Neurosci 2007

Compte et al Cereb Cortex 2000

Calibrate MT network:normalization by strong recurrent

inhibitionInput coming from V1 MT model output

Snowden, Treue, Erickson, Andersen 1991; Heeger 1992

+

+

+

no-attention

+

+

+

attention

2sec

PFC

MT

55Hz

t

Att

No-Att

S

A

Neural label

Attno Att

Network activity pattern shows selective enhancement

Neural label

Multiplicative gain modulation of tuning curve

Att: nonprefAtt: pref

Att: at 90

Martinez-Trujillo and Treue (Current Biol 2004)

MT activity when A=S are covaried

Martinez-Trujillo and Treue, 2004

Attno Att

Network pattern in a single trial

Neural label

R(,S,A)=G(A-) R0(-S)

Modulation ratio

Martinez-Trujillo and Treue, 2004

example MT cell

Population data

R(,S,A)=R(-S,-A,A-S)

R(-S,-A)=G(A-) R0(-S)

G(A-)= 1+g0 cos(A-)

Feature-similarity

Multiplicative scaling of single cell’s tuning curve But selective enhancement of population activity profile in single trials

+

+

+

PFC

MT

2sec

55Hz

+

+

+

Without attentionTransparent motion with attention

Network Activity pattern

R(,S,A)=G(A-) R0(-S)Attno Att

Neural label

Feature-similarity gain principleholds for transparent motion

Att

no Attprediction

Neural label

R(-S,-A)=G(A-) R0(-S)

45Hz

Biased competition