HC 12 Flashcards
What is computational modeling?
Use of mathematics, physics and computer science to study behavior of complex systems.
To find mechanistic explanations of how the nervous system processes information to give rise to cognitive function and behaviour.
A model contains numerous variables that characterize how the brain works. By adjusting these variables and observing the results, model outcome can be matched to behaviour => behaviour is imitated or reproduced in alternative medium= simulation.
Why should we build models?
-Explicit hypotheses and assumptions necessary to test theories of cognition. How does the computer represent and process information?
-Provides framework for intergrating knowledge from various fields
-Allows to observe complex interactions among hypotheses
-Provides ultimate control
-Leads to empirical predictions
-Arteficial lesioning possible to test a model’s validity
What is a neural network?
Computional model inspired by the way biological networks in the human brain process information.
Different levels of complexity: from single neurons & synapses up to abstract connectionist-type or population level descriptions of neural networks.
What is the basic form of neural network?
The basic form has 3 different layers:
-Input (receive information)
-Hidden (processing information)
-Output (transmitting information)
What is the deep neural network?
Neural network with multiple layers between the input and output layers.
What kind of contraints does the biological plausibility of a neural network have?
-Multilayered neural architecture
-Each processing node (neuron) is connected to many others
-Each connection (synapse) is characterized by a connection weight positive values, degrees of excitation and negative values, degrees of inhibition
-Each stimulated neuron receives multiple inputs (dendrites)
-If sum of inputs > threshold: output triggered from receiving node
-Memory, learning depends heavily on changing connection weight => experience-dependent plasticity mechanisms
What is supervised learning?
Develop predictive model based on both input and desired output data weight are adjusted until output gives desired value. This is called classification/regression.
The idea is, training data can be generalized, a model can be used on new data.
The data is labeled and the output is known.
What is unsupervised learning?
Group and interpret data based only on input data. Learn a model that might have generated that set. This is clustering based on pattern.
Data is not labeled, output is not known.
What is reinforcement learning?
Value of output is unknown, but network provides feedback whether the output is right or wrong. This is sequential decision problem.
Provides outcome until correct.
How do the learnings work in short?
Supervised learning= input: data with labels, output: mapping.
Unsupervised learning= input: data without labels, output: classes
Reinforcement learning= input: states and actions, output: state/action
What are biophysical models?
Simulate behavior of neurons/neural network using biologically inspired mathematic equations. Based on specific assumptions and neurophysiological processes.
Based on nodes, can we stimulate a particular node? Clinical application is to test whether we can change specific behaviour.
