Neural representation of objects in the inferior temporal cortex

Symposium held at the European Brain and Behaviour Society meeting, September 17, 2007

Organizers: Pietro Berkes & Yasser Roudi

Inferior Temporal (IT) cortex plays a major role in visual object recognition, and thus is crucial for visual processing in the brain. A critical first step in understanding the underlying computational principles in IT is understanding how visual information is coded by neuronal activity. Outstanding questions include: How are multiple objects represented? What is the relationship between neural activity and perception? Are shapes that are perceived to be similar encoded by similar activity patterns? To what extent are neuronal responses invariant to changes in object attributes and shape? How is invariant representation learned?

These questions are difficult to answer because they depend crucially on hard to control factors such as previous visual experience and degree of attention. Nevertheless, over the past several years there has been considerable progress in answering them. In this symposium we present and review recent experimental and theoretical advances in thie area.

Programme:

9-9:30 Edmund Rolls Invariant object recognition in the ventral visual system
9:30-10 Rufin Vogels Representation of perceived shape similarity in macaque inferior temporal cortex
10-10:30 James DiCarlo The role of visual experience in supporting invariant visual object representations in primate Inferior Temporal cortex
10:30-11 Laurenz Wiskott Is slowness a learning principle of the visual system?

Abstracts:

Edmund T. Rolls
University of Oxford, Department of Experimental Psychology.
Papers available at www.cns.ox.ac.uk

Invariant object recognition in the ventral visual system

In the primate temporal cortical visual areas, the representation of objects is frequently invariant with respect to position, size and even view. The distributed neuronal representation of object identity uses encoding based on the number of spikes, with little contribution of stimulus-dependent synchrony, and with almost independent information conveyed by different single neurons, so that the encoding capacity of the system is very high. A multistage feed-forward architecture with convergence and competition at each stage is able to learn invariant representations of objects including faces by use of a Hebbian synaptic modification rule which incorporates a short memory trace (0.5 s) of preceding activity. This trace rule enables the network to learn the properties of objects which are spatio-temporally invariant over this time scale. A new learning principle utilises continuous spatial transformations to compute invariant representations. It has been found that in complex natural scenes, the receptive fields of inferior temporal cortex neurons shrink to approximately the size of an object, and are centred on or close to the fovea. It is proposed that this provides a solution to reading the output of the ventral visual system, for primarily the object that is close to the fovea is represented by inferior temporal visual cortex neuronal activity. The effect is captured in models that use competition to weight the representation towards what is at the fovea. Some inferior temporal cortex neurons in these conditions have asymmetric receptive fields about the fovea, so that the location of the face with respect to the fovea, and multiple faces, can be represented in a scene. The model has been extended to account for covert attentional effects such as finding the location of a target object in a complex scene, by incorporating modules to represent the dorsal visual system, backprojections, and short term memory networks in the prefrontal cortex to keep active the representation of the object of attention, and does not require temporal synchronization to implement binding. The model has also been extended to a theory of how invariant global motion such as rotation is computed in the dorsal visual system.
Rolls,E.T. (2008) Memory, Attention, and Decision-Making: A Unifying Computational neuroscience Approach. Oxford University Press: Oxford.


Rufin Vogels
Neuro- and psychophysiology lab, K.U.Leuven Medical School

Representation of perceived shape similarity in macaque inferior temporal cortex


James J. DiCarlo
McGovern Institute for Brain Research, Dept. of Brain and Cognitive Sciences, MIT, http://web.mit.edu/dicarlo-lab/

The role of visual experience in supporting invariant visual object representations in primate Inferior Temporal cortex

Although object recognition is fundamental to our behavior and seemingly effortless, it is a remarkably challenging computational problem. Our goal is a mechanistic understanding of how the primate brain accomplished this remarkable feat. Specifically we seek to understand how sensory input is transformed from an initial neuronal population representation (essentially a photograph on the retina), to a new, remarkably powerful form of population representation - one that can directly support object recognition. We are currently focused on patterns of neuronal activity in the highest levels of the ventral visual stream (primate inferior temporal cortex, IT) that may directly underlie recognition. Understanding the creation of the IT representation by transformations carried out along the ventral visual processing stream is the key to understanding visual recognition. In this talk, I will review our results on the spatial and temporal ability of the IT population representation for supporting position- and scale-tolerant recognition. Although several mechanistic hypotheses may explain the remarkable tolerance properties of the IT representation, one of the most intriguing and unexplored is the possibility that visual experience plays an important role in developing such tolerance. I will present results from our ongoing studies aimed at testing this hypothesis using neurophysiology, human psychophysics, and monkey fMRI. These studies illuminate the role of the IT representation in supporting visual object recognition, and provide new constraints on the mechanisms that might produce that representation. Our goal is to use this understanding to inspire artificial vision systems, to aid the development of visual prosthetics, to provide guidance to molecular approaches to repair lost brain function, and to obtain deep insight into how the brain represents sensory information in a way that is highly suited for cognition and action.


Laurenz Wiskott
Institute for Theoretical Biology, Humboldt-University Berlin, http://itb.biologie.hu-berlin.de/~wiskott/

Is slowness a learning principle of the visual system?

Different representations of our visual environment vary on different time scales. Retinal responses vary quickly because they are very sensitive to saccades or object motion while representations in the inferior temporal cortex (IT) display a large degree of invariance and therefore vary more slowly. Turning this argument around leads to slowness as a learning principle. By learning input-output functions that generate slowly varying output signals, units become invariant to frequently occurring transformations, such as translation, rotation, or illumination changes. We argue that this is an effective mechanism by which IT could learn its invariant representations. Interestingly the same principle also leads to a number of complex-cell receptive-field properties even though invariance does not seem to be such an issue so early in the visual system. Some of the simulation results presented here are complemented by analytical results obtained with variational calculus.