c112 (a) When using a feed-forward neural network to solve a… c112(a) When using a feed-forward neural network to solve a classification problem, one alternative is to interpret the network’s outputs as posterior probabilities of class membership, and then to use these posterior probabilities to make classification decisions. Alternatively, we can treat the network as amake the classification decisions directly. Discuss the relative merits of these two approaches. [8 marks] (b) Explain the concept of a likelihood function, and the principle of maximum likelihood in model building and inference from data. [2 marks] (c) Explain the key ideas of a Hopfield neural network for content-addressable, associative memory. In explaining how memories are stored and retrieved, be sure to define the notions of: configuration space connectivity matrix stable attractor basin of attraction network capacity, and its dependence on the number of “neurones” [10 marks] 8 CST.2002.7.9 8 Computer Systems Modelling Consider an M/M/1 queue and represent the state of the queue by the number of customers present. (a) Draw a state diagram for the Markov chain describing the state of the queue showing the possible states and transition rates. Briefly explain the diagram. [4 marks] (b) What is the condition for the existence of a steady-state equilibrium distribution pk (k = 0, 1, 2, . . .) for the number present? [2 marks] (c) Determine the steady-state average number of customers present. [4 marks] (d) Use Little’s law to determine the steady-state average response time that a customer spends in the system. [2 marks] (e) Suppose that a single communication channel is used to carry data items sent by various sources connected to the channel. Assume that each source generates a stream of data items with inter-arrival times which are exponential at rate 2 items/second and that all sources are statistically independent. All the items wait in a single queue and are transmitted one at a time. The transmission times are exponentially distributed with mean 25 milliseconds and are statistically independent. Determine the largest number of sources that can be connected to the channel according to each of the following two criteria: (i) The channel is not saturated. (ii) The steady-state average response time for an item must not exceed 100 milliseconds. (a) A disc is a finite, planar, circular object. Describe an algorithm to findthe point of intersection of an arbitrary ray with an arbitrary disc in threedimensions. Ensure that you describe the parameters used to define both theray and the disc. [6 marks](b) Given the above algorithm and an algorithm to find the intersection of anarbitrary ray with a finite-length open cylinder, a programmer has two choicesfor implementing an algorithm to find the intersection with a finite-lengthclosed cylinder. She could simply use the finite-length open cylinder primitiveand two disc primitives. Alternatively she could implement the finite-lengthclosed cylinder as a primitive in its own right by adding extra code to the opencylinder algorithm. Compare the two alternatives in terms of efficiency andaccuracy. [4 marks](c) Describe the situations in which it is sensible to use a winged-edged datastructure to represent a polygon mesh and, conversely, the situations in which awinged-edged data structure is not a sensible option for representing a polygonmesh. What is the minimum information which is required to successfully drawa polygon mesh using Gouraud shading? [4 marks](d) Derive the formula of and sketch a graph of N3,3(t), the third of the quadraticB-spline basis functions, for the knot vector [ 0 0 0 1 3 3 4 5 5 5 ]. [6 marks]10 HCI(a) Describe two quantitative and two qualitative techniques for analysing theusability of a software product. [4 marks](b) Compare the costs and benefits of the quantitative techniques. [6 marks](c) Compare the costs and benefits of the qualitative techniques. [6 marks](d) If restricted to a single one of these techniques when designing a new onlinebanking system, which would you choose and why? [4 marks]10CST.2002.7.1111 Natural Language ProcessingAn open-domain question-answering system should be able to answer questions suchas the following by searching a database of documents and returning an appropriatetext snippet. For example:User: What debts did Quintex leave?System 1: About $1.4 billion (Australian)extracted from:Quintex Australia Ltd. and Quintex Ltd. together have debt of about $1.4billion (Australian), according to two analysts at Australian brokerage firms.and:System 2: Around ADollars 1.5bn (Pounds 680m)from:Quintex group collapsed yesterday. The failure left corporate debts of aroundADollars 1.5bn (Pounds 680m) and additional personal debts.(a) What natural language processing techniques would be needed to build sucha system? [8 marks](b) What problems would arise as a consequence of the open-domain requirement?[4 marks](c) What problems would arise selecting an appropriate text snippet from amatching sentence? [4 marks](d) How feasible would it be to have the system reliably rank such answers (in thiscase preferring the second)?(a) State and explain (without proving) two different theorems about signalencoding that both illustrate the following principle: strict bandlimiting (eitherlowpass or bandpass) of a continuous signal reduces the information that itcontains from potentially infinite to a finite discrete set of data, and allowsexact reconstruction of the signal from just a sparse set of sample values.For both of your examples, explain what the sample data are, and whybandlimiting a signal has such a dramatic effect on the amount of informationrequired to represent it completely. Write program that dynamically allocates an array large enough to hold a user-defined number of test scores. Once all the scores are entered, the array should be passed to a function that sorts them in ascending order. Another function should be called that calculates the average score. how to delete a subtree whose root is the node p; and return the element of p in python programming by creating a function _delete_subtree()I have the following methodsdef is_root(self, p):”””Return True if node p represents the root of the tree.”””return self.root() == pdef is_leaf(self, p):”””Return True if node p does not have any children.”””return self.num_children(p) == 0def is_empty(self):”””Return True if the tree is empty.”””return len(self) == 0def depth(self, p):”””Return the number of levels separating node p from the root.”””if self.is_root(p): return 0else: return 1 + self.depth(self.parent(p))def _height2(self, p): # time is linear in size of subtree”””Return the height of the subtree rooted at node p.”””if self.is_leaf(p): return 0else: return 1 + max(self._height2(c) for c in self.children(p))def height(self, p=None):”””Return the height of the subtree rooted at node p.If p is None, return the height of the entire tree.”””if p is None: p = self.root()return self._height2(p) # start _height2 recursiondef _iter_(self):”””Generate an iteration of the tree’s elements.”””for p in self.nodes(): # use same order as nodes() yield p.element() # but yield each elementdef nodes(self):”””Generate an iteration of the tree’s nodes.”””return self.preorder() # return entire preorder iterationdef preorder(self):”””Generate a preorder iteration of nodes in the tree.”””if not self.is_empty(): for p in self._subtree_preorder(self.root()): # start recursion yield pdef _subtree_preorder(self, p):”””Generate a preorder iteration of nodes in subtree rooted at p.”””yield p # visit p before its subtreesfor c in self.children(p): # for each child c for other in self._subtree_preorder(c): # do preorder of c’s subtree yield other # yielding each to our callerdef postorder(self):”””Generate a postorder iteration of nodes in the tree.”””if not self.is_empty(): for p in self._subtree_postorder(self.root()): # start recursion yield pdef _subtree_postorder(self, p):”””Generate a postorder iteration of nodes in subtree rooted at p.”””for c in self.children(p): # for each child c for other in self._subtree_postorder(c): # do postorder of c’s subtree yield other # yielding each to our calleryield p # visit p after its subtreesdef breadthfirst(self):”””Generate a breadth-first iteration of the nodes of the tree.”””if not self.is_empty(): fringe = LinkedQueue() # known nodes not yet yielded fringe.enqueue(self.root()) # starting with the root while not fringe.is_empty(): p = fringe.dequeue() # remove from front of the queue yield p # report this node for c in self.children(p): fringe.enqueue(c) # add children to back of queuedef _len_(self):”””Return the total number of elements in the tree.”””return self._sizedef root(self):”””Return the root node of the tree (or None if tree is empty).”””return self._rootdef parent(self, p):”””Return the node of p’s parent (or None if p is root).”””return p._parentdef children(self, p):”””Return the node iterator of p’s children.”””for nd in p._children: yield nddef num_children(self, p):”””Return the number of children of node p.”””return len(p._children) #simply retrun the size of children list. def is_leaf(self, p):”””Return True if Position p does not have any children.”””return self.num_children(p) == 0def _add_root(self, e):”””Place element e at the root of an empty tree and return new node.Raise ValueError if tree nonempty.”””if self._root is not None: raise ValueError(‘Root exists’)self._size = 1self._root = self._Node(e)return self._rootdef _add_to_left(self, p, e):”””Create a new child with element e, and add to the left side children of node p.Return the new node.”””self._size += 1p._children.insert(0,self._Node(e, p)) # add node at the front of the children listreturn p._children[0]def _add_to_right(self, p, e):”””Create a new child with element e, and add to the right side children of node p.Return the new node.”””self._size += 1p._children.append(self._Node(e, p)) # add node at the front of the children listreturn p._children[-1]def _add_by_index(self, p, e, idx):”””Create a new child with element e, and add to the children of node p, indexed at idx.Return the new node.”””self._size += 1p._children.insert(idx, self._Node(e, p)) # add node at the idx node in the children listreturn p._children[idx]def _attach_children(self,p,s):”””attach childre from s to p1 = self._add_to_right(p, s.pop(0))while s[0] != ‘}’: if s[0] == ‘{‘: s.pop(0) #pop out ‘{‘ self._attach_children(p1,s) else: p1 = self._add_to_right(p, s.pop(0))s.pop(0) #pop ‘}’ def create(self,s):”””Create a tree from a sequence, e.g. s =[‘B’,'{‘,’A’,’D’,'{‘,’C’,’E’,’}’,’F’,’}’]”””self._add_root(s.pop(0))s.pop(0) #pop out ‘{‘self._attach_children(self._root,s) #pass in [‘A’,’D’,'{‘,’C’,’E’,’}’,’F’,’}’] def print(self):”””print the tree in preoder with {} to separate children”””self._print_subtree_preorder(self.root())print() Write program that recursively calculates n factorial (n!) a) Main should handle all input and output b) Create a function that accepts a number n and returns n factorial c) Demonstrate the function with sample input from console.Write program that calculates the factorial value of a number entered by the user. 1. Your program should check the value input by the user to ensure it is validWrite a program (array.cpp) that read positive floating point numbers into an array of capacity 100. The program will then ask for one of three letters(A, M or S). if the user inputs something other than one of these letters, then the program should ask for that input until an A, M, or S is entered.Write program that reads in an array of user given size n and then it reads several increment values terminated by 0. The program calls each time a function incarray() that accepts an array A, its size S and an increment value inc and updates yje array by incrementing all its values by the passed value of inc. (a) State the typing rule for ML let-expressions, let x = M in M0 , using typing judgements of the form A, G ` M : t where A is a finite set of type variables, G is a finite function from variables to type schemes and t is a type. [3 marks] (b) Give an example to show that in a let-expression let x = M in M0 , the letbound variable x can occur polymorphically in the body M0 . Give the proof of any valid typing judgement that you use. [5 marks] (c) Give the ML typing rules for the unit value (), for reference creation ref M, for dereferencing !M, and for assignment M := M0 . [4 marks] (d) Explain how the combination of the typing rules from parts (a) and (c) leads to unsoundness of the type system. How does the revised definition of ML modify the typing rule for let-expressions in order to restore type soundness? You are required to design a Sentient Computing System using Active Bats to locate people and physical objects. (a) Indicate how you would model the environment and represent space in your architecture. [8 marks] (b) Show how you would use the tracked objects’ dynamics to process and filter incoming location data. [8 marks] (c) Discuss the feasibility of a spatial monitor that generates events in terms of geometric containment and overlapping. [4 marks] 15 Additional Topics VNC is a protocol for remote access to graphical user interfaces. (a) Describe the VNC architecture. [6 marks] (b) Indicate what steps are taken to minimise network traffic. [5 marks] (c) How does the protocol adapt to networks and clients of different speeds? [5 marks] (d) Discuss how suitable VNC might be for used in third-generation mobile telephony (3G) systems. Computer Science Engineering & Technology C++ Programming COMMUNICAT 1020

Don't use plagiarized sources. Get Your Custom Essay on

Top answer: c112 (a) When using a feed-forward neural network to solve a…

Just from $10/Page