Lecture 2 Flashcards
Two types of navigation
Path Integration
e.g. desert ants: once food is found, go back to nest via much more direct path. Kept memory of motions and created a vector. Seem to remember how many steps they take (changing leg lengths causes under/overshoot). Combination of heading (direction) + distance travelled.
Local/Cognitive Map
e.g. pigeons: use external cues e.g. landmarks - must have internal representation
1) Tends to accumulate error whereas 2) is highly accurate.
1) Works between locations, 2) doesn’t
Types of cue we use to navigate (X2)
Ideothetic cues (aka egocentric): internally generated, provide basis for navigation by path integration. E.g. vestibular input, proprioceptive information. Used subconsciously to compute turns/distance travelled, only works in known environment, lots of error.
External landmark cues (aka allocentric). Also only works in known environment.
Animals use both.
Supervised vs unsupervised learning.
Tolman (x2 experiments).
In computational terms, associative learning is “supervised”. System monitors the prediction error and adjusts accordingly.
Unsupervised learning occurs when there is no explicit learning outcome e.g. spatial learning (as we walk around, we learn about the environment even if there is no reward for doing so).
“Toll”man: two groups of rats in a maze. First group got food reward, second group didn’t until day 11. On day 11, second group did better (less deviation from the path). Were learning unsupervised, weren’t just following same path every day like group 1. Better cog. map of maze.
Also showed rats make detours. If you put a block in the maze, rats DON’T try the last successful path (use experience of map to guide their behaviour - argues against behaviourism, supports existence of cog. map) (see images)
Radial arm maze (2 types of memory)
Central platform (where animal is placed) with arms radiating out containing food. Visible cues around maze (for orientation). Optimal strategy is to visit each arm only once.
REFERENCE memory: used BETWEEN trials (e.g. memory of which arms are baited)
WORKING memory: used WITHIN trials (e.g. memory of which arms have already been visited).
Morris Water Maze
Rats must find submerged platform. Must learn relationship between platform position and fixed cues outside the pool.
“cued” versus “spatial” versions of the tasks
Involvement of hippocampus
Radial arm maze and morris water maze can also have cued versions. Cues are found within the apparatus itself so doesn’t require spatial memory (dissociates hippocampus from other structures).
Hippocampus important for spatial but not cued learning (cued tasks performed normally with any hippocampal lesion. All lesions disrupt spatial learning. Evidence that hippocampus = site of cog. map)
Hippocampal lesion sites (X4)
“Entor”hinal cortex: major cortical input to hippocampus
Septum: major subcortical input
Fornix: input/output route for subcortical –> hippocampal projections
Fimbria: structure that carries CA3 commissural connections to ipsilateral CA1
Place cells and place fields.
Neural basis for the cog. map. Found within hippocampus (more evidence for map being in hippocampus). Fire APs when rat enters the cell’s “place field” (where the cell fires maximally). Can record from live animals using electrodes and use camera to track position of animal.
A sufficient number of place cells can map a given environment. NOT topographic.
Presumed to be pyramidal cells.
Place fields are very stable (up to 6 months in rat) and resistant to rotation of the environment. Robust spatial map.
How spatial maps are constructed (X3 hypotheses)
COG. MAP HYPOTHESIS: construction of map relies on external sensory cues (doesn’t depend on motion cues or memory). Place cells provide good neural basis for this.
PATH INTEGRATION HYPOTHESIS: path integration can provide enough info to generate map. Relies on internally generated cues provided by motor activity (direction, velocity etc.). Requires that basic map is present that can be “customized” to current environment. Place cells provide poor neural basis for this.
MEMORY SYSTEM HYPOTHESIS: place cells form part of a more general “episodic” memory system. Different aspects of any event encoded as episodic memory.
If the hippocampus can’t support path integration, what can? (4 types of cell)
HD (head direction) cells: found in the presubiculum, fire when animal faces particular direction. Fire in 90 degree arc around preferred direction. Preferred direction is independent of environment. May “orient” spatial maps in hippocampus.
Grid cells (in medial entorhinal cortex): fire in regular grid-like pattern which forms equilateral triangles. Adjacent cells map adjacent points (TOPOGRAPHIC). Grid cells are contextindependent and maintain constance spacing between peaks in firing rate - so can map DISTANCE.
Border cells: active when walls are encountered.
Speed cells: indicate how fast animal is moving.
