3 3 [25 marks) (a) Describe one process for creating vector representations (5 marks) (embeddings) of words, as used by Word Vec. In your answer make clear reference to how words are represented and to the role played by neural network leaning. You can focus your answer on either skip-gram or CBOW. (d) Consider a simple artificial neural network (such as a simple [5 marks] Perceptron). In your own words, describe how the weights are modified during the process of learning a simple function such as binary AND, OR OR. Compare and contrast the learning that occurs when the output is correct and when the output is Not correct (c) Describe in your own words how Reinforcement Learning (RL) [5 marks] can learn to efficiently explore a problem space such as a maze. The diagram below displays three different trajectories across the search area, while the +1 and -1 indicate the final reward/punishment associated with each trajectory. Focus your answer on how RL can overcome the delayed feedback (or sparse feedback) problem. You may focus your answer on learning a Fixed Policy (FP). 1 >-1 (d) Describe how you would adapt the K-Nearest Neighbours (KNN) [10 marks] algorithm to estimate the sale value of a second-hand (pre- owned) car, given a database of recent sales of pre-owned (such as discussed in class). Identify 5 attributes recorded for each car- sale that you would use, as a basis for identify similar cars from this database. Marks will be awarded for considering different attribute types, and for the range of values that can be usefully compared. Use example values to illustrate your