Homeostasis

Question 1

What is homeostasis

Answer 1

It is the process of when a system returns to a set point

Question 2

define homoeostasis in terms of neural networking

Answer 2

Using gradient descent on an error surface to reduce the discrepency between the output and the target until the difference between the two is 0.

Question 3

How to use the error surface to make a system reach homeostasis

Answer 3

A Braitenberg vehicle such as m= s1- s2 can be wired up to perform homeostasis.
1. Provide the vehicle with a target connected to 1 sensor and an output which is connected to the other sensor.
x= target - output
2. providing one of the wires is negative.
3. Plot the error function of the vehicle.
error(x)= x2
The error surface is all the possible states the vehicle can be in.
4. Put in the value of x (x = target - output)
5. perform gradient descent on the function
m= -2 (target - output)
This shows that the motor will stop when the target - output = 0, homoeostasis.

Question 4

What real life phenomenons can be explained by homeostasis?

Answer 4

• The thermostat in a house.
  The thermometer detects the temperature, is provided with a target (20degrees), the motor is adding hot or cold water to heat or cool the house until the discrepency between the target and output = 0.

Question 5

How is this the first steps to the rescorla wagner model

Answer 5

The rescorla- wagner model is supervised learning
Supervised learning involves presenting a network with examples of input patterns along with information about what the correct outputs should be in response to those inputs. For example, the delta rule involves presenting a neuron with sensory inputs as well as target output value, and iteratively modifying the synaptic weights / wires.
The rescorla wagner model and the delta rule add on to homeostasis because they involve adjusting the weights/ wires to reduce the error.