Result for Neural Network Task 1

The results obtained from the MLP networks with different neuron tried are shown in the Table below. After 20,000 iteration of training, it could be seen that the best result obtained was the first one using 10 hidden neurons with score 42%, although the RMS error was the highest.




A screenshot of its RMS error from NeuralWork is shown in Figure 3.1 below.

Knowledge-Based System Task Four

The first part of this task was to run a agent-based system as shown in the above figure using flex agent. Every agent was connected up and registered with the facilitator agent (dvlafac). The whole idea of this agent-based system was to allow the client agent (myagent) to extract information of a particular vehicle from the individual registration office through the facilitator. The client agent could also add information on to the registration office. Four KQML messages were sent out by myagent to test out the system. They are shown in the table below with respective message codes and performatives used.

The second part of this task was to write relations to the newoffice agent to perform validity checking on a registration mark. There are two main relation for performing this checks, they are validate(Regmark) which will perform validation and validhere(Regment) which will interpret whether the vehicle was registered in Glasgow. The codes in Task KBS1 and Task KBS2 was converted to relations and reused in this task. The codes in the file GET_MARK.KSL were modified such that the questions and write statements were all taken out. The program was tested in the console and run successfully. The agent was started and connected to the network. Some messages were sent by myagent to test out the system.