Lecture 10 Flashcards
Neural network theory
An explicit attempt to create a non-computational model of how cognition works and learns.
Non-language-like as cognition at its fundamental level is not language but brain-like.
Signal strength
How much signal is traveling from A to B, analogous to the firing rate between neurons/groups of neurons.
Valence
Exhibition/Inhibition,
Weight of connectivity
Connections are not symmetrical in capacity to hit each other.
Ex. Amygdala impacts neocortex. The neocortex analyzes situations where the amygdala signals radically wrong situations (urgent/priority).
Grief
The amygdala wonders where is someone, neocortex analyzes and reminds they are gone.
Slowing the amygdala may hinder grief response but not end the grief.
Learning via neural networks
Mathematical equations through patterns of activity that propagate through the network.
Input and output of nodes increase the level of signals sent to other nodes or inhibit firing.
Submarine analogy
To detect the difference between a rock and a mine, a submarine takes an image and converts it into a digital signal.
Statistical Analyzes
Target value - Granted value = Error value
Back-propagation of error
To analyze how each connection is statistically responsible for the error, you can alter them by taking the error and propagating it back through the network. It will alter the connections based on probabilistic data.
Variance errors
The network can pick up the wrong thing due to the learning processing occurring in a certain context.
Homoncular
It is a form of supervised plasticity hence supervised learning, not allowing to generalize brain-like functions.
Neural networks simultaneity
The network can run multiple simultaneous processes at once, it is a parallel processor.
Hinton: Sleep-awake cycle
Unsupervised learning where it should be getting information from the world without direct feedback.
Sleep-awake cycle: stages
- Compression of input figures out invariant across data.
- It uses compressed data to try and generate variations and particularizes data.
Compresses- generates - sensitivity selection.
Perceptions: Minsky
Perceptions can not run certain cognitive functions.
Older brain areas have less opportunity to be exapted therefore they involve fewer functions.
