CSCI 4850-5850 - Neural Networks | 1. (2 pts) What class of problems are single-layer networks incapable of solving? 2. (2 pts) If a multi-layer network uses a linear activation function for the hidden layer units, what class of problems can it solve? 3. (2 pts) What general class of activation function do the hidden units in a multi-layer network need to for them to solve nonlinear problems? 4. (2 pts) Which set of weights in a multi-layer network cannot be learning/changed/updated using the standard delta rule? 5. (2 pts) If target vector information is not available for updating hidden layer weights in a multi-layer network, what other general learning approach could be used to update those weights instead? 6. (2 pts) Why is the generalized delta rule not a good model for how biological neurons update their synaptic connections? 7. (2 pts) When using error backpropagation, what are the names of the two mathematical variables/terms which are multiplied together to calculate the change/update for a connection weight? 8. (2 pts) When using the generalized delta rule, what is the name of the first step or forward pass of the training process? 9. (2 pts) When using the generalized delta rule, what is the name of the second step or backward pass of the training process? 10. (2 pts) What type of function is the generalized delta rule trying to minimize