Edmund T. Rolls, Oxford Centre for Computational Neuroscience
Oxford University Press (2012)
pp. 65- visual perception
Summary and review of the above section
Keywords: visual perception, ventral stream, dorsal stream, reward/punishers
Visual perception: It is pointed out that computer vision science has not yet mastered the problem of perception. Single neurons in the primary visual cortex respond to edges or bars at particular angles or orientations on the retina. The problem is to get from there to neurons in the inferior temporal cortex that selectively respond to faces or objects regardless of their position on the retina. The ventral stream projects from the primary through the secondary visual cortex and V4 to the inferior temporal. This stream is mainly visual with other modalities lacking significant inputs into this area. Perception here is seen as being mainly concerned with what an object is. The representation here is also value free in that it is not related to the reward/punisher qualities of the representations. This stream is mainly concerned with building a representation of the objects.
After the formation of the visual representation, it is projected to where it can form associations with other modalities. The ventral stream converges with other unimodal streams for taste, smell, touch and hearing in a number of areas and particularly in the amygdala and the orbitofrontal. These are also areas involved with evaluating rewards and punishers. The amygdala and the orbitofrontal project via the hypothalamus to the autonomic or involuntary nervous system controlling functions such as heart rate, breathing and digestion and also the endrocrinal or hormonal system. They also project to the ventral striatum, which relates to action and behaviour. Further to the orbitofrontal, the cingulate cortex learns the associations between behaviours and rewarding or punishing outcomes. The learning process may be one of trial-and-error. Eventually behaviour may become a matter of habit, but the trial-and-error approach can reactivate if the environment changes and the habitual response is no longer satisfactory.
Reward-related learning takes place on several levels. The autonomic or involuntary system can learn or become conditioned to unconscious processes such as salivation or increased heart rate in response to pleasant or unpleasant stimuli. Separate from the autonomic is a system of stimulus-response learning involving the striatum and other parts of the basal ganglia; this system can lead to the formation of habits, which subsequently do not require, or can even be disrupted by conscious thought.
Brain processing is here divided into three teirs. In the visual system the inferior temporal cortex is in the first teir in producing a conscious representation. In the second tier, the reward value of these representations becomes conscious in regions such as the orbitofrontal and the cingulate cortex. These evaluations involve a convergence on the orbitofrontal from the visual, somatosensory and other sensory areas.
The third tier is where choices are made between the presented values. The cingulate cortex is involved in goal-based learning of actions to get rewards (instrumental learning). This involves learning the connection between behaviour and reinforcers (reward or punisher outcomes). This may involve trial-and-error learning. Such learnt actions may transform into a habit, but if the habit stops delivering the expected reward, the trail-and-error approach may return. Neurons in the orbitofrontal allow the system to correct itself. These neurons become activated if the outcome of behaviour is different from what has been predicted. Rational thought is seen as a fourth tier in decision-making, particularly when it comes to deferring rewards in order to obtain longer-term benefits.
In a small part of the cortex each neuron does a calculation, and the GABA activity of inhibitory neurons screens out the weaker neuron responses leaving the relatively sparse distribution of the stronger responding neurons. Computations start off in networks that are relatively separate, such as the ventral and the dorsal stream. In regions such as CA3 in the hippocampus and also the inferior temporal visual cortex, closely connected neurons contribute to a single calculation. The spatial position of a neuron within the brain is itself seen as a code, so a neuron in the inferior temporal cortex can code for an image, but the same type of neuron in the auditory cortex codes for a sound.
In teir 1, the identity of stimuli is represented independently of reward value. This is argued to allow us to gather information about an object relative to the environment without immediate reference to whether the object is rewarding or punishing. This evaluation comes later. The subsequent reward value of stimuli is coded into the firing of neurons. The level of neuronal activity directly correlates to the subjective assessment of pleasantness. This is partly a question of what subjects are paying attention to. If they pay attention to pleasantness, they will generate more activity correlated to pleasantness, and if they pay attention to intensity, they will have more activity correlated to this. The response to the same stimuli can also change when a food approaches satiety. Different single neurons respond to different stimuli from visual, olfactory and somatosensory areas. The orbitofrontal also represents negative reinforcers, Further, some orbitofrontal neurons respond only when an expected reward is not obtained. The orbitofrontal has a common scale for the value of different types of reward which is more flexible than always choosing the same type of reward.
When a choice has to be made between the reward value of stimuli, the orbitofrontal represents the value of rewards, and does not represent behaviour or action. The medial prefrontal is just in front of the orbitofrontal. The cost of obtaining the reward has to be subtracted from the reward value indicated by the orbitofrontal. Stimuli of different value rather than actions or behaviour are represented in the orbitofrontal. The orbitofrontal projects to the anterior cingulate which represents both reward/punishers and the costs of obtaining/avoiding these. The orbitofrontal represents only the values and not the costs of rewards. The anterior cingulate is also indicated to project to the mid-cingulate cortex, which monitors the outcome of actions. The orbitofrontal additionally monitors the non-delivery of expected rewards. The orbitofrontal and amygdala also project to the basal ganglia and particularly to the ventral striatum region of the basal ganglia, which directs actions and behaviour on the basis of reward/punisher and other assessments.
The whole system relies on a common neural currency with which to evaluate different stimuli and future costs. Stimuli that are masked or available for too short a time to enter consciousness can still influence emotional assessment, although the experience of the emotion itself is conscious. The evaluation system may also be modulated by cognitive processing centred on the dorsolateral prefrontal, while the rational processing of the dorsolateral may be modulated by the evaluation process, which may bias attention towards certain rewards in the environment. Activation of the orbitofrontal as seen in neuroimaging are directly correlated with subjective reports of the pleasantness of the stimuli. This ‘like something’ is not a general property of neural networks, and seems suggestive of a special type of processing. The basal ganglia receive inputs from most parts of the cortex, but do not feedback to most of it, suggesting that they exist to execute the results of projections from the rest of the brain. The suggestion is that computation in the cortex finds its output in the basal ganglai, which then project to the motor cortex to create behaviour.Tags: Edmund Rolls, visual perception Posted by