Fast read-out of object identity from macaque inferior temporal cortex

Chou P. Hung*, Gabriel Kreiman*, Tomaso Poggio, James J. DiCarlo

1. Sample of spiking activity and local field potentials in macaque inferior temporal cortex

Sample of spiking responses
Sampel of LFP responses

2. Unsupervised clustering of neuronal responses yields similar categories

Correlation coefficient matrix for population responses to two pictures
Clustering of correlation coefficient matrix
Clustering of correlation coefficient matrix with different numbers of clusters (n=2, 3, …, 10)

3. What is a category?

Random group assignment
Classifier performance versus similarity
Confusion matrix
Extrapolation to other pictures within a category

4 Read out of scale and position

Classifier performance for reading out scale and position
Latency of scale and position read-out
Overlap in high SNR sites for group read-out, scale read-out and location read-out

5 Categorization and identification

Read out performance for categorization and identification
Latency of categorization and identification
Latency of categorization and identification, invariance
There was a strong overlap between the best neurons for the categorization and identification tasks
There was a strong correlation in the categorization SNR and the identification SNR

6 Spike sorting

Spike sorting and classifier performance
Separation of single units and multi-units

7 Codes

A family of codes parametrized by the spike count window
Single bins
Bursts of spikes

8 Robustness

Robustness to neuronal drop-out during testing
Robustness to failures in information transmission

9 Wiring specificity

Wiring specificity

11 Neuronally plausible decoding

Neuronally plausible statistical classifier and comparison of different statistical classifiers

12 Invariance to scale and location

Robustness of the classifier to scale and location changes
Invariance for novel images
Novel objects images
Invariance performance as a function of the number of sites
Invariance for C2 units of the standard model

13 Detection of stimulus onset

Detection of stimulus onset

14 Classification from simple image properties

Stimulus categorization from simple image properties

15 Dependence on the number of sites

Classifier performance versus number of sites with fits
Number of neurons required to discriminate a number of objects at a fixed performance

16 Active versus passive viewing

Discriminability for active versus passive viewing

 

 
Top