2. Derive the gradient descent training rule assuming that the target function representation is: a) Define explicitly the cost/error function E, assuming that a set of training examples D is provided, where each training example d e Dis associated with the target output td. b) Consider the instance space consisting of integer points in the x, y plane, where 0Sx, ys 10, and the set of hypothesis consisting of rectangles (i.e., being of the form (a x b, cy d), where 0 3 a, b, c, d s 10). What is the smallest number of training examples one needs to provide so that the CANDIDATE-ELIMINATION algorithm perfectly learns a particular target concept (e.g., (2x4,6y 39)? Explain your answer in a clear manner (i.e., explain when can we say that the target concept is exactly learned in the case of the CANDIDATE-ELIMINATION algorithm and what is the optimal query strategy). What is the difference between the Best First Search and the Beam Search algorithms? c) Which types of knowledge can we distinguish in Case-Based Reasoning (CBR)? Provide a short explanation of each of the types. d) The four parts carry, respectively, 30%, 25%, 25% , 20% of the marks