Task 4 Flashcards
What are assumptions of neural networks?
- neurons integrate information
- neurons pass on information about input levels
- brain structures are layered
- infleunce of I on J depends on connection strength
- learning = changing weight (strength of connections9
What are cognitive processes?
results of computations in parallel by large numbers of neurons
-> information is distributed over many neurons and connections
->
What is connectionist modelling?
- inspired by our brain -> unit = neuron
- demonstrates content addressibility and fault tolerance
- shows typicality effects in retrieval
- not corresponding to human memory
= human memory can be accessed by content
= human memory is organised content-addressable retrieval - two types of models
(= can be used to perform parallel constraint satisfaction)- concerned with local representations
- concerned with distributed representations
What are pattern associators?
- describe how stimuli become linked when they are
repeatedly presented together - classical conditioning
- takes place by modifying the strength of the connections between input units and output units
What is fault tolerance ?
even if some synampses on neuron i are damaged after lerning, netinputi following the presentation of a recall cue may still be a good approximation to the correct value
-> this netinput provides the pattern carrying the recall cue consists of a reasonably large number of axons the correlation will not be greatly affected by a few missing items
-> after passing through the binary threshold function, the result may well be correctly recalled
what are the three phases of competitive learning?
- excitation
> excitation of the output units proceeds in the usual fashion by summing the products of the activity of each input unit and the weights of its connection - competition
> the units compete with each other and the identification of the winner may be achieved by selecting the unity iwth the highest activity value - weight adjustment
> weight adjustment is only made to connections feeding into the winning output unit in order to make it more similar to the input vector (for which it was the winnner)
What is constraint satisfaction?
- When activity flows through a connectionist network in response to an input, each neuron influences the state of all other neurons to which it is connected
- Each input can be perceived as a constraint on the activity state exhibited by the network
- By influencing each other, neurons push and pull to satisfy certain constraints
- The overall state eventually ends up representing a compromise between all of these constraints, satisfying each as much as possible
what is an autoassociator?
- Output reproduces input -> recurrent connections
- Reproduces the pattern that was presented as input by changing weights:
internal input = external input - generalisation
- fault tolerance
What is BCI?
- brain-computer interface (bridge between brain and external device)
- initially focused on helping paralysed people control assistive divices using their thoughts
- can give real-time information about what is going on in the brain
- neural networks = pattern associators
What is BCI used for in the future?
- neurofeedback training tool to improve cognitive performance & to affect altertness
- for managers: ability to monitor and control attention lvels
- smart home: mental stages can be detected and nearby devices can be adjusted
What are risks of BCI?
- can determine user´s atentional levels, but cannot differentiate what the attention is directed to
- 15-30% of individuals are unable to produce brain signals robust enough to operate BCI
- privacy
- ethical questions for workplace use
- high potential for abuse
> can be hacked
How is information stored in connectionism?
- distributed fashion
- processing in parallel = all neurons perfom operations at the same time
- all knowledge in connectionist model is super-imposed on the same set of connections
