Flashcards
Computational Neuroscience L2
Cards in this deck (47)
- 1What is the hippocampus a link between?Limbic system and the cortex
- 2Where is long term and highly contextual memory stored?Associated cortex
- 3What are the primary functions of the hippocampus? (2)Processing of new memory and constructing spatial maps
- 4What does CA stand for? What are the four regions?1) Cornus Ammonis - horn of Ammon
2) CA1-CA4 - 5Describe the pathway through the hippocampus.EC2 to dentate
EC2 and dentate to CA3
CA3 and EC3 to CA1
CA1 to subiculum
Subiculum and CA1 to EC deep - 6What is the DG? What is unique about the DG? What cells does it contain? What do these cells connect to?1) Dentate gyrus
2) One of the few areas in the brain that undergoes adult neurogenesis
3) Granule cells (principle cell type, excitatory) connect to CA3. Also contains inhibitory basket cells - 7Where does the main input to the hippocampus come from? What does this name mean?Entorhinal cortex - 'near the smelling area'
- 8What connection characteristic do excitatory neurons in the DG show? What do they lack? What other network exists here?1) Excitatory feedforward
2) Excitatory collateral connections
3) Recurrent connections between excitatory and inhibitory cells. - 9What is the excitatory network in CA3 referred to as? Why?1) Recurrent network
2) Collateral excitatory connections within CA3 - 10What forms the connections between the DG and CA3 region?Mossy Fibres
- 11What forms the connections between the CA3 and EC region?Perforant pathway
- 12What forms the connection between the EC and the DG?Perforant pathway
- 13Describe the pattern of connectivity throughout the dentate gyrus. How does this lead to regulation?1) Granule cells connected to basket cells
2) Excitatory input is modulated through the inhibitory basket cell network. - 14Describe the structure of connectivity within the hippocampus. What does the loop at CA3 indicated?Recurrent/autoassociative network
- 15What is the pathway which runs between CA3 and CA1?Schaffer collateral fibres
- 16Describe top down models. (2)1) Vague model which expresses an idea about what the whole brain is doing
2) Mathematics to describe what the brain is doing - 17What have patients with hippocampal damage shown about the type of memory the hippocampus is involved in? (3)1) Involved in declarative
2) Long term memories are stored outside the hippocampus
3) The hippocampus is not involved in procedural memory - 18What is an auto-associative memory?Patterns which represent memories along with some dynamic that complete partial patterns.
- 19What does an 'on' neuron refer to? An 'off' neuron?1) Rapid spiking
2) Tonic spiking - 20How is a novel object initially encoded?Neurons activated independently - each neuron encodes some characteristic of the novel object. Sequence of on and off neurons encodes the object
- 21How does recall of an object occur?Network is presented with the partial pattern and evolves into the complete pattern.
- 22How does representation of an object come to be through a single neuron?One neuron receives increased input from those neurons which encode the separate characteristics of the object - the strength of that pattern of input is increased forming an autoassociative input.
- 23What are McCulloch-Pitt neurons?Neurons described as either 'on' or 'off'
- 24What do the binary 'on' or 'off' states refer to biologically?Firing or not firing.
- 25How is a partial pattern completed during recall?The rest of the network is recalled through strengthened connections.
- 26What equation determines the strength of a neuron?wij ie the strength of the connection between neurons i and j.
- 27What does the McCulloch-Pitt model of neurons leave out?1) Details of firing
2) Possibility of different firing rates - 28What letter marks the number of neurons? Activity of neuron? Strength of connection? What assumption about the strength of connection is made?1) N
2) xi (activity of neuron i)
3) Wij
4) Wij=Wji - 29Which equation models the change in synaptic strength? What are the sort of values which can be inputted?Where n=learning rate
a= sparseness of activity of neurons
2) Either 1 or 0 ie neurons i and j are either on or off.
- 30What are the possible mathematical outcomes of the change is plasticity equation?If both neutrons are on - ie xi=1 and xj=1 then Wij=n(1+a^2 -2a)
If both neurons are off - ie xi=0 and xj=0
then Wij=n(a^2)
If one neuron is off and the other is on - ie xi=1 and xj=0
then Wij -= n(a^2)-an - 31What do the mathematical outcomes signify?1) Both neurons on - there is a large increase in strength of connection
2) Both neurons off - small increase in connectivity strength (a is a very small number <1)
3) Only one neuron active - decrease in strength of connectivity - a is less than one so that n(a^2) - 32What size is n generally?n is generally very small however in the hippocampus where new memories must often be learnt very quickly following even just a single presentation of the stimulus n is relatively large.
- 33What happens to the state of neurons during recall?Some of the neurons within the pattern are being held in the activated state (ie they are activated) the rest of the pattern/network evolved according to the threshold input rule - ie if the input from the partial pattern is above threshold the rest of the network will be activated completing recall.
- 34What is the threshold equation? Describe it. What condition must be met for recall to occur?Given input to neuron i is equal to the sum of the strength (Wij) of the inputs (xj). hi must be greater than theta, the threshold of activation for that network (individual to different networks).
- 35What is the equation which describes the activity of a neuron based on its threshold and input.xi = 1f(ri)>thetai
Where ri= input into neuron. If the sum of input is not greater than theta (threshold for causing activity) then neuron i will not fire (xi=0) - 36Explain how this results in complete threshold and a scenario where it wouldn't.If nodes {0,5} are activated following the learning of pattern {0,2,5}, within which the connections will have been strengthened, the input to neuron 2, h2=w12+w52 will be larger than threshold and subsequent dynamics will switch neuron 2 on completing recall. If a different pattern were activated within this network then the activity would die as hi would all be subthreshold as there has been no strengthening of connections.
- 37Describe the concept of interference within a neuronal network.Neurons may be involved in multiple patterns which cause connections within one pattern to strengthened whereas connections in another pattern may need to be weakened - if the amounts of strengthening and weakening of these connections within the different patterns are similar then this may impede retrieval and prevent that neuron participating effectively in either pattern.
- 38When is pattern overlap more likely to be possible and still provide functional recall?In a large network - but too much overlap will prevent retrieval.
- 39What is capacity (ie the amount of information) which can be stored proportional to?The number of neurons, N.
- 40Describe the 'hand-waving' argument which determines the number of patterns (ie memories) which can be stored.Number of neurons = N
Therefore number of connections (roughly) is = N^2
Amount of information in a pattern is N (ie the number of neurons in that pattern each encoding a characteristic)
Therefore the number of patterns which can be stored is equal to N^2/N which is equal to N. - 41What assumption is made in the 'hand-waving' argument? What happens when this assumption isn't met?That each pattern is independent. If they aren't all independent then capacity is reduced.
- 42What happens if patterns share fragments/sub patterns? What can this lead to?1) Sub pattern connections can become very strong ie between {0,1}
2) This can lead to erroneous recall/pattern activation following activation of subpatterns,
- 43—Different possible erroneous pattern completion following strengthening {0,1} due to subpatterns.
- 44What are auto-associative networks capable of storing? What does this mean they have never been proven useful for?1) Random patterns
2) Machine learning - 45What is the suggested solution to the problem of sub-pattern erroneous completion.Solved through EC-DG-CA3 pathway. Dentate gyrus randomises the connectivity between EC and CA3.
- 46Describe how the neurons in the EC-DG-CA3 pathway are matched during learning? What is this process called and what is the benefit of this?1) EC and CA3 are matched through DG. (EC to DG) and (DG to EC) are randomly connected. CA3 is an autoassociative memory store and DG, a feed-forward network, passes the information on to EC but randomises it on the way
2) Pattern separation - helps to keep similar patterns separate. - 47—X
Information
- Description
- Year 2 Techniques in Neuroscience Flashcards on Computational Neuroscience L2, created by Anna Hogarth on 01/05/2016.
- Created by
- Anna Hogarth
- Views
- 2
- Rating