Simulating Language 2015, pre-reading 5 questions Question Title * 1. Which of the following statements is true of neural networks? Learning operates by changing connection weights They consist of sets of connected units Every unit must be connected to every other unit in the network Activation spreads between units along weighted connections Each connection has a weight The incoming activation to a unit determines whether that unit will be activated Learning operates by changing activations Question Title * 2. The matrices we have been using to model production/reception matrices can be re-described as networks. How? Each meaning and signal is represented by a single unit, activation spreads between meanings and signals along weighted connections. For production, all signal units receiving incoming activation exceeding a set threshold become active; for reception, all meaning units receiving incoming activation exceeding a set threshold become active. Each meaning and signal is represented by a single unit, activation spreads between meanings and signals along weighted connections. For production, the signal unit receiving the most incoming activation becomes active; for reception, the meaning unit receiving the most incoming activation becomes active. Each meaning and signal is represented by a single unit, activation spreads between meanings and signals along weighted connections. For production, all signal units receiving any incoming activation become active; for reception, all meaning units receiving any incoming activation become active. Question Title * 3. Hebbian learning, as described in the reading, involves which of the following operations? Weakening weights between units which are active at the same time Weakening weights between units which have differing activations (i.e. one is on, one is off) Strengthening weights between units which have differing activations (i.e. one is on, one is off) Strengthening weights between units which are active at the same time Question Title * 4. [Some of you have trouble with images in these surveys, so I am writing this question without them! I am going to write the weights out as a grid: the first two shows the weights from unit m1 to signals s1 and s2, the second row shows the weights from m2 to signals s1 and s2]Imagine a network representing an agent with two meanings and two signals. Initially, all connection weights are 0, i.e. it looks like this:0 00 0This learner makes four observations of the signalling behaviour of some other individual: meaning 1 is conveyed using signal 2; meaning 2 is conveyed using signal 1; meaning 1 is conveyed using signal 1; meaning 1 is conveyed using signal 2. Assuming the learner is applying Hebbian learning, what will the connection weights in the network be after learning? -2 10 3 2 1 0 1 0 00 0 1 21 0 Question Title * 5. Is there anything you would like me to know before the lecture - any stuff you want me to go over again, anything you struggled with, any worries you have? Done