# View-invariant object representation in anterior and posterior inferotemporal cortex: A machine learning approach

## TL;DR

This study uses machine learning to compare how different parts of the inferotemporal cortex represent objects from different viewing angles.

## Contribution

The novel contribution is using machine learning to show stronger view-invariant object representation in the anterior IT cortex (TE) compared to the posterior IT cortex (TEO).

## Key findings

- Area TE showed viewing angle tolerance of 30–90° for objects with prior learning of different views.
- Area TEO showed viewing angle tolerance of only 30°.
- Late time period activity in TE suggests stronger view-invariance representation.

## Abstract

Inferotemporal (IT) cortex is the final visual area in the ventral stream where object information is processed. Previous electrophysiological studies showed viewing angle tolerance of 30–60° of single IT cells to the objects experienced in discrimination at each of several viewing angles, and to the objects experienced in learning association of different views. IT is divided into anterior (cytoarchitectonic area TE) and posterior (TEO) parts. It was reported that single cells in area TE showed the viewing angle tolerance while those in area TEO did not. In the present study population activities were compared between cell populations in area TE and those in area TEO using machine learning algorithm. An object set consisted of four similar objects created by deforming a prototype object, and four views each separated by 30°. A population vector was created by aligning responses of the cells to each object image. A classifier was trained by support vector machine (SVM) to create a hyperplane that separated one object from the other three objects at the same viewing angles, and then tested by response vectors to the object images at different viewing angles. In area TE, dynamics of the performance evaluated by d’ showed viewing angle tolerance of 30–90° to the objects with prior experience in learning association of different views. In area TEO, populations of the cells showed the viewing angle tolerance of 30°. Significant increase of the d’ values in area TE in the late time period for the objects with prior experience in learning association of different views may suggest view-invariance is more represented in late time period than early time period. These results suggest that viewpoint invariance is expressed more strongly in the TE region, and expressed in part in the population of the TEO cells.