Greg Detre
Wednesday, 24 May, 2000
Prof. Rolls - B&B IV
inferior convexity � infero-temporal
object working memory
working memory
loop � recurrent collaterals
why doesn�t memory wander? what keeps the firing within a select sub-population?
strong synapses only onto other neurons coding for the same area of space
(attractor/auto-association networks)
hence: the importance of getting the right synaptic weights
Ral paper:
first, required a working memory component (object)
then made to remember a particular location
asked: whether it�s the same neurons in STM for objects vs location (vs speech chunks etc.)
found that: many neurons coded for both � not absolute separation
7 items in STM probably limits objects + location
no of separate active at once
normally remember order of items
Rolls theory: links up the separate components� attractor networks
why can�t we remember more? � interference would collapse into conglomerated state
evidence for separate localised attractor nets for each item? Goldman-Rakic lesions (& Fonahashin)
prefrontal unitary function?
dorsal � all working memory
conditional spatial response � image-mapping of remembered image � motor space
prefrontal � visual representations of objects and motor maps
orbitofrontal
not attractor � apparently different
respond to visual stimulus if associated with reward � but they reverse very quickly (one-trial learning)
they remember the most recent reward to a visual stimulus
= the memory of the reward value of a visual stimulus
unitary functioning � remembering over time intervals
but can reload very quickly
\ = a unitary function in the sense that it�s all concerned with recent memories
but very different internal workings
but it is still part of the prefrontal � it receives from the dorsomedial thalamus too (but the medial magnocellular bit of it instead)
Fuster
overarching function = true for dorsal
orbito � just an inhibitory role? he underplays it
pioneer: showed delay-related activity
no supervisory attention centre or central exec (Shallice, Baddeley)
the general unified-seeming phenomenon of attention is formed just by the separate attention areas together
like any area of cortex, the STM needs to be capable of ignoring external stimuli, and take over control of behaviour
�there doesn�t need to be a single exec � control can be allocated according to task demand
how that allocation happens is the difficult part of attention
perhaps: the basal ganglia might be where the selection takes place of which processing becomes behaviour
projects back to the cortex, so it could then turn down the gains from the inputs from the other areas not currently in control of behaviour
frontal lobes
frontal eye fields = almost part of the prefrontal
motor areas (just in front of the central sulcus)
auditory depends on back propagation from the language centres
auditory = also easier because there are fewer parameters???
how do we control the receptive field gains for one location rather than another?
response-outcome learning: reward information from the orbitofronal � dorsolateral prefrontal
STM = helpful/necessary, because rewards often aren�t immediate
what about long-term response-outcome learning?
monkeys are very bad at it, because having to wait for long-term rewards is very rare
the only way that humans are able to do it is via a totally different mechanism, i.e. use language to represent contingencies
Petrides � studied PET scanning in humans
decided that: dorsolateral = critical in non-spatial ordered memory tasks
seems inconsistent with (his) other findings � perhaps because he used the statistical minimum of 3 monkeys
hunger
brain mechanisms which control appetite and motivation
reading � just the starred ones
How is food intake
controlled?
what�s
the difference between prefrontal and frontal?
prefrontal is a small area within frontal
where
is the orbitofrontal � what�s it normally called?
is a
lobe solid or cortical (i.e. surface)?
= surface
�the
visual regions of TE�
what�s the difference between the active item in STM vs the others being held?
by allocating according to task demand at a specific point, as you have to, e.g. arbitrating between the inputs to the basal ganglia, you are effectively adopting a single exec sort of system, aren�t you?