Answer the following questions with regard to commonly applied neural network models.

Question 2 (2+3+3+3+4=15 points) Answer the following questions with regard to commonly applied neural network models. 1. The Oja rule solves a serious problem of the Hebb rule. Explain what the problem is and how it is solved. [2 marks] 2. A student working on a French letter recognition project has segmented a large number of letter images. He then applied the images to train a neural network model. After convergence, he visualized the connection weights as shown in the following figure. eeeeB8SSSSEEFIE DOC SEEUP Okode DO NO Coro odggico ocuCUDOLU 5099 8 Oda EN 7 DuluddjaffNUXXIT niifiddd AMXLII nuRAHniLL inuitliiiii Briefly describe the neural network model which the student applied and discuss the possible ways to use this result. [3 marks) 3. What is the important role of the sigmoid function in back-propagation learning? [3 marks] 4. Explain the "momentum term" that is often included in the back-propagation learn- ing algorithm. [3 marks)