Spatial and Cognitive Maps II

Question 1

Place cells in linear track

Answer 1

Place cell response can also be measured in a linear track
Respond at one location in track - 1d, x
Their responses to deformation (shortening) can be measured
Cells firing at all locations so can represent space
Position on track represented = not actual distance
Order of firing the same even if compress representation

Question 2

linear track in another dimension

Answer 2

Sound dimension
Press lever and hold, increasing sound as press
No spatial navigation tho
One get to certain frequency = get reward and release lever
Travelling in sound space
Similar to traversing a 1d track in physical space

Question 3

tuning to location in sound sequence

Answer 3

Hippocampal cells are often tuned to specific location in sound sequence
Similar to tuning linear track
Recording in hippo, tuning to single location around sound axis
Tile whole of sound
Cells in ca1

Question 4

what would be a control experiment to show that a cell is not just responding to sound frequency

Answer 4

How do we know its not tuned to diff frequencies
= change sound
Change contingencies - not being rewarded, not pressing level or reward, just hearing sound

Question 5

cell responses are…

Answer 5

Dependent on context

Question 6

descrive passive playback - what does place cells do

Answer 6

Pace cell - like neuron does not respond if same sound freq profile is played passively outside of task
No activation, so suggests it is task - 1d navigation
So does have to do wirh navigating envir

Question 7

tuning correspondence across spaces

Answer 7

Record animal in task and put in box - have them navigate in box
Cell 1= put animal in box while navigation and see it looks like place cell
Cel 2 = wont always have a response in both tasks
Some cells showing tuning to single location in sound sequence
As well as place tuning in random exploration
Cells in CA1
But if have response in both tasks = looks like place cell
Single lcoation of space - not structured or repeated intervals tho
Correspondence bewteen place cell in classic sense - sort of single lcoation responses in 1d navigation

Question 8

grid cells across cognitive spaces

Answer 8

Etr cortex
Some cells show tuning to multiple locations in sound sequence as well as grid tuning in random exploration
See grid cells - several peaks, repeated intervals, repeated structure in some task - respond to multiple locations
Complicated structure
Representing something common irrespective of space

Question 9

grid cell spacing

Answer 9

Varies with anatomical lcoation
Diff grid cells tile space with diff lattices with diff orientations and diff scales
Peaks closer = finer details
Peaks far = represent bigger space

Question 10

spacing correspondence across spaces

Answer 10

Navigation vs sound task = grid cells
Grid cells with wider fields in spatial envir also have wider fields in sound task = suggest shared neural mechanisms
3 sets of spacing distribution - matches with sound task
Module 1 = tends to have smaller fields in sound task, modules 2/3 = wider
Look how far peaks are but in time (for sound task)
= GRID cells in both tasks- precise grid cells in navigation

Question 11

shared neural mechanisms for maps

Answer 11

Key properties of tuning across cells
Structure and size of features
More general- involved in many tasks
Place cells - at location, shared
Grid cells = structures rep, share key properties

COMMON circuit mechanism in hippocampal-entorhinal system are used to represent diverse behavioural tasks, possibly supporting cognitive processes beyond spatial navigation

Question 12

cogntive maps across dimensions

Answer 12

Spatial —> cognitive maps
Apply to abstract thinking
Navigate space that is not pshycial but probabuly still has simailr attributes
Not just analogy - it’s actually encoding diff spaces in same way

Question 13

identifying grid like representations in humans

Answer 13

Hard, deep in brain
Can leverage structure fo grid fields (symmetric) to identify areas that have grid like rep in navigation
Use fmri - task= navigate vr world in scanner
:. Method identifies etr cortex and mpfc
= have grid like reps

Question 14

navigating an abstract space

Answer 14

Navigation in abstract spaces
We can navigate 2d space of bird shapes - where leg and neck length varies (smooth transition)
Or can have completely random abstract obejcts and have to learn transition bewteen objects
Simailr to navigation In phsycial or virtual 2d space

Question 15

similar areas are activated while navigating tehse spaces

Answer 15

Navigation in the 2 spaces = identifies etr and mpfc in fmri exps as having grid like representations
Suggests navigation system is not just used for navigation in physical envir but it is a more general system for navigating spaces that ahve relational structure
Irrespective of type of space you navigate - any type of space with relational structure between things

Question 16

transitive inference requires what

Answer 16

A map

Question 17

describe how transitive inference requires a map

Answer 17

Need general map
Get snippets of transition then put them together in a coherent space = allows you to knwo structure of space as a whole
Maps allow you to do transitive inferences - have to build global rep
= build space from single interactions

Breakdown of experience, inference and choice at test time in spatial navigation and transitive inference relies on chaining together separately observed sequences of observations

Question 18

testing transitivity in preferences = set up

Answer 18

Pairs of stimuli presented together - one stimulus rewarded per pair
Test on pairs not experienced to test whether ordered presentation is being represented

Question 19

testing transitivity in preferences =results

Answer 19

Lesion Entorhinal cortices or the fornix (output of hippocampus) vs control group (lesion parietal cortex)
Deficit in pair requiring computing transitivity
- accurate on middle pairs, all lesions do well
- deficit in b vs d relational probe = need to ahve map to know b>c and c>d, animals with lesion in navigation system = perform at chance level and lack ability to build map and relational structure
No deficit in end pair test = a vs e and in random pairing (wx and yz- no one knows what to do )
:. These areas needed to build ordered representations - map

Question 20

what is needed for transitivity in preferences

Answer 20

hippocampus and navigation system

Question 21

reprsenting task structure

Answer 21

Many Hippocampal neurons exhibit spatially specific firing, firing of place cells encodes animals position in envir
Ex 1 = maze, spatial navigation in same way = shares task structure with ex 2
Ex 2= nose poke smell to diff outcomes, initiate responses - same then separates Depending on which chosen
Basically equivalent but think about them differently

Question 22

can both types of representations co exist??

Answer 22

Can ahve purely spatial or task representation
- should think about them both when look at decision
Is it in pshycial space or more abstract transitions between events

Question 23

learning structure over tasks

Answer 23

Harlows task = diff instantiations of task share common underlying structure - only one object is rewarded, images chnage tho
= can be exploited to facilitate faster learning
Over multiple exposures to task = monkeys acquire underlying structure - learning set, and use it to learn faster in new instantiations - learns over tine = then will be more accurate and pick object quickly (takes only 1 trial to know which object to pick)
Learning higher level abstract representation of task

Question 24

Ofc Lesions= explain

Answer 24

Deficit in tracking changing reward probabilities
If lesion = can’t switch
Monkeys can also learn to track changing probabilities of reward but ofc lesions cause deficit = can track changes in reward unless best option
Lesions broad here so large effects

Question 25

inability to associate choice and reward after ofc lesion

Answer 25

How are task rewards affecting choose Each trial = independent - look at trial history and compute what current estimate of value of diff options is and should pick best one In healthy animal = use previous trials to guide choices Ofc lesion = not using choice history as much- no longer knwo which recent choice caused which recent reward

Question 26

what is perturbed after ofc lesion

Answer 26

Map of state space Reduced state diagram Ofc lesion = unable to pair reward with choice that causes it Correct state space if required for correct credit assignment Like left leads to one reward, right leads to another separate reward =normal Ofc lesion = left and right both lead to SAME reward - lose ability to separate rewards so credit goes equally to both wells, can’t back track

Question 27

what can help performance

Answer 27

Knowledge of task structure Behaviour can adapt in one shot as on experience of a small/large reward is enough to know that lock identity has switched - so have high level map of task (bc no trial and error) Da and rpe and identifying cognitive map Expect 2 get 4 = good, expect 6 get 4 = bad - can shift - dont have to do trial and error In envir = get small or big reward or some intermediate If get big reward envir and then get small reward = know envir changed immediately Lick a lot if high reward, lick a little if low reward envir = immediately know that envir switched bc have block structure

Question 28

what are da responses consistent with

Answer 28

ALSO see evidence of maps in da firing Shift expectation as fast as shifting behaviour Firing of da neurons consistent with map of task space - model based Animal has expectation to be in high or low regard block and rpe scales accordingly

Question 29

map based vs graph based representations

Answer 29

In spatial domains= Euclidean based - metric based = 2d xy rep, grid like cells, smooth thing we can traverse in any direction, many contexts, more continuous Graph based = more graph, more structured, not as smooth transition Can exist simultaneously- can switch between the 2 In non spatial domains = Knowledge is map based, info endowed in terms of continuous dimensions (social group = represented in terms of personality characteristics) Graph based = encoded in terms of distinct links between items (social group represented in terms of social connections in group) Not sure if combo exists here

Question 30

spatial representations

Answer 30

Evidence in humans

Question 31

cognitive maps across dimensions

Answer 31

same cells in navigating other types of spaces Ideas of spatial or cognitive maps= allows you to understand some responses/behaviours more - states and locations in space

Question 32

testing transitivity in preferences = exp probes

Answer 32

A>B>C>D>E B vs d = test of transitivity A vs e =non transitive novel pairing - bc e never rewarded = know a is better always bc is rewarded, not same case for other pairs

Question 33

conclusion

Answer 33

when navigating naturalistic envir = range of cell representations found in cognitive map of hippocampus-entorhinal system Same representations in coding abstract or conceptual spaces = suggest single coding mechanism underlies both physical and conceptual spaces -understanding how states of world relate to each other