0lRetrieved from: introduction_to_ai_consolidatedUGѬ,aCsE |N |`cDc|cTc.@|Ѭ,ap ||@|@|rL >|r,aѬC:\Documents and Settings\jsmith.CEMS\Local Settings\History\desktop.iniF dv ׃Awvf53  3Isu}nZ:f53g untitledzE$|A~= 1qA~s|=T s|<A~X |Ȏ$E |N |؎$Č,A||l<|p ||o>|b>|hTXGΌ,XZXNCX(ЌE CXE |N |T|h.@|Ќp ||@|@|0>| >|c:\RXXZЌc:\Respondus Projects\ai_questions_respondus |X   |Xl C#w(w,<279+<= AqGnOV\}bYhlVrtv~})((Aqɱi9+p*9utw+rk `ra"C%9(+d14C8,<?A#E M_QfY\^`djtmqIu&y}i89Osm=uHlV6LWhich of the followiMMost of these methods are available to you. If you have some specific questions about the content or tutorial work, please ask them via the discussion boards that every one can benefit from them. Remember that if you call in without an appointment, it is highly unlikely that we wil be able to see you, even to make an appointment.eAccessing the lecture notes made available in the "course documents" section of UWE Online.kAccessing supplementary materials made available in the "course materials" section of UWE Online.By using the class discussion boards available via UWE Online. For example, looking for related questions, asking questions, or starting new discussion threads.By discussing things with your group, either in the tutorials, orpossibly using the group discussion boards so that you can do this off-line and also have a record.MBy discussing the issue with one of the tutors duriong the tutorial sessions.WBy emailing the tutor to arrange an appointment to discuss things with them in private.Calling by the tutor's office and knocking on their door in the hope that they will have some spare time now, or will make an appointment.??????Which of the following choices are available to you to get help and more information about this course? Note that more than one answer may be correct.This is an example oSorry, try again.Artificial Intelligence approaches based on theories of how the mind works. Forexample how the mind works with a representation of the world, and reasons with facts.V

This is part of "good old-fashioned AI" and is covered in semester 1.

Approaches to building Artificial Intelligence systems based on models of how the brain works - for example artificial neural networks.yANN's arean important part of what is now known as "Computational Intelligene" and are covered in semester 2.Artificial Intelligence approaches based on using the principle of evolution to "breed" solutions to problems that cannot be tackled by conventional means.}So-called Evolutionary Computation is another important part of Computational Intelligence and will be covered in semester 2.Spiritual and religious issues concerning what it means to be human and their placement within an overal ethical/moral framework.^In semester 2 we will be discussing the notion of what is "intelligence". In semester 2 we will look at models of computation based on Darwin's principle of evolution and modern genetics. However, we will not be discussing religious issues, or whether evolution or some external being is the cause of humanity's presence on earth. Well done.This is an example of a multiple choice question in which only one answer is correct. Which of the following are not part of the syllabus for this course?4This is a trivial e

Sorry, try again.

Hint: first do the same in decimal, then convert your answer to a binary value - i.e. a string of 0s and 1s.

@?Q

10 + 6 = 16.

to represent this in binary you need 10000 i.e. 5 bits.

This is a trivial example of a numerical question.

Typically I will allow you some leeway - fiver percent either way.

How many bits of binary code do you need to represent the answer to the sum 10+6 ?

6iFill in the two blan Try again.StrongWeak Well done.Sorry, try again. Well done.sorry, try again. Well done.Fill in the two blanks in the following sentence.

[Strong] Artificial Intelligence is the attempt to produce machines that exhibit general purpose intelligence. In most cases this involves building a model of the world around them, and applying reasoning in order to plan and make predictions so as to achieve desired goals. In contrast to this, [Weak] Artificial Intelligence is the attempt to build machines that appear to exhibit intelligent behaviour, usuallywithin a particular limited sphere.0zComplete the follwoi no sorry.Strongmodel reasoninggoalsweakappear Well done.,Complete the follwoing sentence by making appropriate selections from the choices presented.

[Strong] Artificial Intelligence is the attempt to produce machines that exhibit general purpose intelligence. In most cases this involves building a [model] of the world around them, and applying [reasoning] in order to plan and make predictions so as to achieve desired [goals]. In contrast to this, [weak] Artificial Intelligence is the attempt to build machines that [appear] to exhibit intelligent behaviour, usually within a particular limited sphere.6Weak] methods are thSorry. Hint these are not the same as "weak" or "strong" philosophies of what might be possible with Artificial Intelligence.WeakStrong Well done.Sorry, try again.

Well done.

Sorry, try again. Well done.

[Weak] methods are those that apply a general problem-solvingapproach but do not rely on any knowledge of their environment, or the particular task.

[Strong] methods, in contrast, make direct use of problem or task specific information.

Which of these statesorry, try again.To pass the Turing test a machine must be able to fool a human interrogator each time, for an unlimited period of interrogation.wSorry. That would be asking quite a lot. After all an interrogator could simply believe that everything was a machine.To pass the Turing test a machine must be able to fool a human interrogator most times, for an unlimited period of interrogation.
GSorry, wrong. Hint: could you guarantee an uninterrupted power supply?To pass the Turing test a machine must be able to fool a human interrogator each time, for an limited period of interrogation.
wSorry. That would be asking quite a lot. After all an interrogator could simply believe that everything was a machine.To pass the Turing test a machine must be able to fool a human interrogator most times, for an limited period of interrogation.Correct. In fact Turing's predictions of what would be possible by the year 2000 are quite low - five minute of interrogation and 30% success rate. Well done."Which of these statements is true?You are taking the plTry again. Hint: all but one of these have been used to find out previous attempts to pass the Turing test.;Questions concerning the emotional state of the respondent.YQuestions with dual meanings, where a human might be expected to guess the right context.fQuestions designed to generate an irrational response in a human - e.g. rude or repetitive questions.@Questions asking the respondent to perform complex calculations.Questions in chinese.???? Well done.You are taking the part of the interrogator in a Turing test. You are typing questions into a terminal and have to decide whether the responses are coming from a person or a machine. Which of these lines of enquiry do you think might be helpful?Searle's uses his ChSorry, incorrect. Do you think that his choice of the chinese language is relevant to the argument he is making.Would it make any difference if he used strings of binary symbols?d

Well done. His choice of the chinese language is not relevant to the argument he is making.

Searle's uses his Chinese Room scenario to argue that machines can't pass the Turing test because they cannot understand chinese.

Searle uses his Chinsorry.Well done. In fact there are severalexamples of machines that have fooled quite a few people by adopting certain stategies which exploit human weaknesses. Eliza is the best known example.Searle uses his Chinese Room scenario to argue that a machine could pass the Turing test, even if it could not understand chinese. Therefore the Turing test is not a good test for intelligence.Searle uses his Chin 2sorryCorrect.Searle uses his Chinese Room scenario to argue that even if a machine acted in a way that appeared to be intelligent, it would not really be so because it would not understand what it was doing.Searle is suggestingSorry. His point is that just because a machine appears to exhibit intelligence, it does not necessarily possess mental states or consciousness that would be required to count as "strong" Artificial Intelligence.Well done. Searle's point is that although to the observer the room appears to understand the questions and reply sensibly, in fact none of the components (the books, the person inside) do.Searle is suggesting that a if machine could pass his Chinese Room test it would be successfully exhibiting "strong" Artificial Intelligence.Which of the followi 2Sorry, try again."Beating a world champion at chess.<Wrong. IBM's Deep Blue computer beat Garry Kasporov in 1997.zProducing a diagnostic system that can be more accurate and reliable that junior doctors for a specific class of ailments.*Sorry, MYCIN did that starting from 1972.WProducing a system that correctly predicted mineral deposits when humans did not.
Sorry, PROSPECTOR did that.it was developed between 1974 and 1983 and in 1980correctly predicted the presence of a mineral deposit worth 100 million dollars..Driving a car for 100km without accidents etc.That's been done. In 1995 a car went from Munich to Cpenhagen on the publich highway with users not being involved 95% of the time. The longest uininterrupted period was 158km. Try googling for "Darpa Grand Challenge" results of the annual trials.Passing the Turing test.Correct. Still some way off.3Inventing something that has been granted a patent.Koza successfully applied for a patent for an electronic circuit that had been invented by Genetic Programmnig in the early 2000s.Correct. Well done.wWhich of the following tasks have not yet been achieved by a machine based on "weak" Artificial Intelligence?You are asked to proSorry, incorrect. Think again about what are the inputs, model and outputs in this case. Which of these three spaces are you being asked to search? This is an Optimisation problem.

Correct.

There are a huge number of possible routes between any two points, and these form the input space.The database of distances respresents the model of the system. The output is the sum of the distances between adjacent points on the route.

You are given the model (database), and the condition on the output (find the shortest route) , so this is an optimisation problem.

This is a Simulation problem.

Sorry, incorrect.

Think again about what are the inputs, model and outputs in this case.

Which of these three spaces are you being asked to search?

This is a Modelling problem.

Sorry, incorrect.

Think again about what are the inputs, model and outputs in this case.

Which of these three spaces are you being asked to search?

This is a trivial problem.VAre you sure? Try thinking about how many possible route there might be to examine ...+Well done. This is a typical problem where we have a model (in this case a set of junctions and distances between them) and a set of possible inputs- routes between two specified end-points. We are given a criteria that our input should meet -namely that the distance (output) should be minimised.You are asked to produce a program that provides the shortest route between two points in the UK. You are given a database with the distances between every road junction. Which of one the following statements is true?Doctor X has collectSorry, incorrect. Try again. Think about what the inputs, outputs and model might be in this case. Are the outputs discrete or continuous? This is an optimisation problem.3This is a classification type of modelling problem./This is a regression type of modelling problem.This is a simulation problem.Well done. This is a typical problem where we have a set of inputs (measurements) with corresponding outputs (diagnoses). The task isto search the space of possible models for one thatmatches known inputs to known outputs, andcorrectly predicts outputs for unseen inputs. As phrased above,the output comes from a finite set:malignant or benign. This is different to the alternative of coming from acontinuous scale, such as a probability of being malignant. Thefore it is a classification rather than a regression problem.Doctor X has collected a database of measurements of skin lesions such as size, colour, circumference, measure of smoothness, etc. The database also contains the the biopsy result (malignant or benign) for each lesion. You are asked to produce a diagnostic assistance system. When given the measurements from a newly presented skin lesion, this should predict whether it is malignant or benign. What kind of problem is this?You are put in a mazsorry.An optimisation problem. Well done.A simulation problem.ano. Think about whether you are searching the space of inputs, models or outputs to your problem.+A classification type of modelling problem.no. Think about whether you are searching the space of inputs, models or outputs to your problem. Just because it might be helpful to build up a model of the maze in your head, does that means you care about how faithful all of that model is?'A regression type of modelling problem.no. Think about whether you are searching the space of inputs, models or outputs to your problem. Just because it might be helpful to build up a model of the maze in your head, does that means you care about how faithful all of that model is?Well done. Another way of stating the problem is to find the shortest sequence of moves that reaches the maze exit. this makes it clearer that we are looking in the space of inputs to our path-following algorithm.=You are put in a maze and told that you have to try and find the shortest way out. Each step can be one of {north, south, east, west} and all steps are 1m in length, unless they would make you collide with the maze walls. In that case your step-count is increased but you do not move. What type of problem is this?0Complete the followi_sorry, try again. These are fairly crucial notions for the whole of computational intelligence.d enumerationdiscretestrongsearch solutionsgloballocal neighboursoptimapWell done. These are important notions for considering learning as search, and thinking about what search means.Complete the following sentence by making appropriate selections from the choices presented.

Learning can be thought of as [search] through a space of possible [solutions] to find one that meets a goal state, or which has the highest value. [global] optimisation is the term we use when we are trying to find the solution whose value is higher than all others. In contrast, [local] optimisation is the term we use to describe methods that find soltions that are better than all their [neighbours]. For some search spaces the notion of neighbourhood is natural, but for others the number of local [optima] present may depend on how we define the neighbourhood of a solution.6Fill in the blanks isorry, try again.search solutionsGloballocal optima,peaks Well done.)no, I want a precise 6 letter word here . Well done.+Sorry, hint: the answer is a 9 letter word. Well done.Sorry. Well done.5 letter word, not global. Well done.Sorry.TWell done.You have clearly understood this bit well if you got it right first time.

Fill in the blanks in the following sentence.

Learning can be thought of as [search] through a space of possible [solutions] to find one that meets a goal state, or which has the highest value.[Global] optimisation is the term we use when we are trying to find the solution whose value is higher than all others.In contrast, [local] optimisation is the term weuse to describe methods that find soltions that are better than all their neighbours.For some search spaces the notion of neighbourhood is natural, but for others the numberof local [optima,peaks] present may depend on how we define the neighbourhood of a solution.

If we take altitudedSorry, try again. Was your choice higher than its surroundings? Was there a higher point somewhere?The summit of Everest.+The summit of Dundry hill in South Bristol.Bristol docks at high tide.Bristol docks at low tide.ZPeasedown St John. For non-locals, this village lies on a flat-topped ridge south of Bath.xWell done. A local optima that is better than its surrounding points, but not necessarily better than all other points.If we take altitude to be equivalent to quality, and use the natural distance measure to define closeness, we can think of the surface of the earth as a search landscape. Which of the following points is a local, but not a global optima?If we take altitude 2Sorry, try again.The summit of Everest.+Correct - there is nowhere higher on earth.(The top of Dundry hill in South Bristol.WSorry. It may be higher than it's surroundings, but is there somewhere higher on earth?Bristol docks at high tide.xNo. Remember we have defined quality to be the same as distance above sea level, and so we are looking for a high point.Bristol docks at low tide.xNo. Remember we have defined quality to be the same as distance above sea level, and so we are looking for a high point.ZPeasedown St John. For non-locals, this village lies on a flat-topped ridge south of Bath.No. If the ridge is flat-topped, will the village be higher than its neighbours? Remember we have defined quality to be the same as distance above sea level, and so we are looking for a high point.Correct.If we take altitude to be equivalent to quality, and use the natural distance measure to define closeness, we can think of the surface of the earth as a search landscape. Which of the following points is a global optimum?Is it true that a sosorry. If a solution is better than all other points, (globally optimal) then it must be better than its neighbours however you define a neighbourhood.Yes. If a solution is better than all other points, (globally optimal) then it must be better than its neighbours however you define a neighbourhood.MIs it true that a solution which is a global optimum is also a local optimum?A solution which is_No. If the global optimum was a neighbour, then the orignal point would not be a local optimum. Well done.A solution which is a local optimum in a search space may have a neighbour which is the global optimum. Assuming that the neighbour is not the same solution, is this sentence true or false?You are given a data"Sorry, wrong selection. Try again.It is definitely possible to write a program that will come up with a guaranteed shortest path in a time that increases linearly with the number of road junctions.It is probably not possible to write a program that will come up with a guaranteed shortest path in a time that increases linearly with the number of road junctions.It is definitely not possible to write a program that will come up with a guaranteed shortest path in a time that increases linearly with the number of road junctions.*A human solving this problem would look at a map and instinctively see likely routes in a straight (ish) line between the end-points. The ability of a computer to do this will depend on whether it has the co-ordinates of each junction, but it could come up with solutions without that knowledge.??pThe problem is NP. That means there is probably no polynomial time algorithm that can decide if a given route is the shortest. Algorithms do exist which are guaranteed to find the shortest path. However their run-time increases exponentially with the number of junctions. This is good example of where heuristics are needed to come up with "good enough" solutions.You are given a database with the distances between every road junction in the UK. The task is to produce a program that, given a start and end-point somewhere in the UK, provides the shortest route between them.Which of the followng statements are true about this problem?0Complete the follwinsorry, try again.probably impossible definitely polynomially guaranteed exponentiallyrequired Well done.OComplete the follwing sentence by making appropriate selectios from the choices presented.

You are given a database with the distances between every road junction in the UK. The task is to produce a program that, given a start and end-point somewhere in the UK, provides the shortest route between them. Choose the right words to fill in the gaps in this sentence. The problem is NP. That means that it is [probably] [impossible] to write a polynomial time algorithm that can [definitely] decide if a given route is the shortest in a time that increases [polynomially] with the number of junctions. Algorithms exist which are [guaranteed] to find the shortest path. However their run-time increases [exponentially] with the number of junctions. This is good example of where heuristics are [required] to come up with "good enough" solutions.Match the following>Sorry, try again. In one case the clue is in the description./Exhaustive enumeration of all possible answers.

  1. Generate an initial random assignment of the ten binary values - the incumbent solution.
  2. Start from the left hand end and for each bit in turn:
  3. When you get to the right hand end of the solton, return to (2)
  4. Repeat until no further improvements can be found.
Agenetic algorithm with fifty binary coded individuals, using recombination between parents and mutation to create new solutions.Simulated Annealing.Deterministic Search. Local Search.'Population-based Meta-heuristic search.3Meta-heuristic search that accepts worsening moves. Well done.

Match the following search algorithms with their appropriate descriptions.

Assume thateach solution is completely specified by the values of 10 binary variables.

Also assume that it is Wednesday.

6Fill in the blanks i 2sorry.completecannot,can't,does not,doesn'tallcorrect. incorrect.correct.>incorrect. I've allowed four possible versions of what I want.correct.hIncorrect. Remember this is an ideal search method - these things may not always exist for all problems. well done.Fill in the blanks in this description. A search algorithm should ideally have the following properties:

  1. It should be [complete] i.e. be capable of generating every solution, so that no possible solution is overlooked.
  2. It should be nonredundant so that it [cannot,can't,does not,doesn't] generatea solution more than once.
  3. It should be informed so that it only generates solutions that satisfy [all] of the constraints.
Both Depth-first andincorrect. Both of these are blind search methods which search according to the structure of the tree, regardless of the node values. Well done.Both Depth-first and Breadth-first search rely on the presence of heuristics to estimate the distance of a state from the goal state.Which of the followi 3It traverses the tree from left to right, testing each node at a given level before either moving to the next level, or quitting if a solution has been found.*It is guaranteed to find a valid solution.pIt is a fast and efficient method for searching trees where there are a large number of branches from each node.NIt is a quite good at searching trees where the branches have variable depths.???? Well doneFWhich of the following statements are true about breadth-first search?Which of the followi 4Sorry, try again.yIt descends iteratively down each branch of the state-space graph, working form left to right, until it finds a solution.KRegardless of the nature of the graph, it is guaranteed to find a solution.cIf different solutions have different qualities, it is guaranteed to find the one with lowest cost.It has very low storage requirements as it just needs to keep a record of what node it is currently searching, and the position in the tree.??\Well done. depth-first can indeed get trapped if the tree has infinite branches e.g. loops.EWhich of the following statements are true about depth-first search?Which of the followi 5dIncorrect. Does it help if you think of a hillclimber where heuristic = height below global optimum?KIt should never under-estimate the distance from anode to the goal state.IIt should never over-estimate the distance from anode to the goal state.%It should be applicable to all nodes.It should be fast to compute. Well done.qWhich of the following properties does a good heuristic evaluation function not need to possess?The 8-tile problem cSorry, try again.Dh1node = count of the number of tiles in the wrong place.h2(node) = sum of the Euclidean distance of each nodefrom its desired location. In this case the Euclidean distance is the square root of (x2 + y2) where x is the horizontal displacment and y is the vertical displacement.h3(node) = sum of the Manhattan distance of each node from its desired location. In this case the Manhattan distance is the sum of (x + y) where x is the horizontal displacment and y is the vertical displacement.h4(node) = h3(node) + count of tiles that are directly in line with, and out of order with,another from the same final row or column. Well done. 

The 8-tile problem consists of a square board containing 8 numbered pieces and a gap, into which one piece can be slid. The task is to move the tiles about so that you end up with a square with the gap in the middle and the numbers sorted clockwise around the edge i.e. {1, 2, 3} in the first row, {8, gap , 4} in the second row , and {7, 6, 5} in the third, which I will denote as {1, 2, 3, 8, g,4, 7, 8, 6}.

From any given starting position, all others can be viewed as states of the system, can be reached by a sequence of valid moves. In order to inform the searchprocess, we can create various heuristics describing how far each state is from the goal in relaxed versions of the problem. Order the following heuristics according to how they dominate each other.Which of the followi 6 Incorrect.It maintains a list of nodes to search next, which it sorts in order of: f(node) = h(node) + g(node), where h(node) is a heuristic estimate of that node's distance to the goal state, and g(node) is the cost, so far, of reaching that node.It maintains a list of nodes to search next, which it sorts in order of: f(node) = h(node) + g(node), where h(node) and g(node) are two different heuristic estimates of that node's distance to the goal state.CIt is guaranteed to find the shortest path to the optimal solution.`It is guaranteed to find the shortest path to the optimal solution faster than any other method.CIt relies on the provision of a good heuristic evaluation function.???CorrectBWhich of the following statements are true about the A* algorithm?Order the followingDSorry. Think about how many nodes need to be stored at any one time.Depth-first search.Greedy Ascent Hill Climber.Steepest Ascent Hill Climber.Breadth-first Search.Best-First Search. well done.Y

Order the following search algorithms in terms of increasing storage requirements.

4PAccording to wikpediSorry, try again. Remember that although humans currently give birth quite late, this is because modern medicine allows us to live longer, and social/cultural customs may change over time.OAj@6Correct. I assumed an average childbirth age of 20.fAccording to wikpedia, scientists currently think that the split between the genus homo (typical modern example is you) and the genus pan (typical modern example is chimpanzees) happened around 6 million years ago. If the average breeding age is 20, and the theory of evolution is correct, how many generations has it taken for you to evolve since the split?Which of the followi 7Sorry, try again.ZThey will reliably produce a high quality solution in a fairly predictable amount of time.ZThey are guaranteed to produce the best possible solution in a predictable amount of time.WThey are guaranteed to produce a high quality solution in a predictable amount of time.YThey are not guaranteed to produce a solution of any higher quality than random guessing.?? Well done.Which of the following statements are true of well-designed Evolutionary Algorithms, and other Computational Intelligence techniques? Note I am deliberately avoiding precisely specifying the terms “high quality” and “fairly”.Which of the followi 8Sorry, try again.fIt is a means of automatically producing computer programs that should meet a set of desired criteria.{It cannot be considered to be creative since it is still tied to human preconceptions of what the solutions must look like.nAlthough it can produce good solutions to many problems, they may be highly complex, even for simple problems.DIt will always produce solutions which humans can easily understand.CIt is an over-hyped method that will never prove human-competitive.xIt is a promising technology which is already producing human-competitive systems at a fraction of the development cost.??? Well done.EWhich of the following statements are true about Genetic Programming?What features of a g>Sorry, try again, most but not all of these features can help.The use of a population means that it is more likely to have starting points that lie within the basins of attraction for different local optima.YThe use of mutation which can potentially change the value of every gene in an offspring.UThe use of crossover which can combine partial solutions from different local optima.hThe element of randomness in selection which means that the currently fittest solution may be discarded.@The use of a well understood binary representation for problems.????0Well done, all these four are valuable features.}What features of a genetic algorithm mean that it is less likely to get trapped in a local optimum than simple hill-climbing?Holland's Simple GenSorry, try again.IThe best individual in one generation will always be present in the next.SThe best individual in one generation will always be present in the set of parents.KThe best individual in one generation may be present in the set of parents.~If the best individual is picked to be a parent, then it will always be present in the set of offspring produced by crossover.If the best individual is picked to be a parent, then it will always be present in the set of offspring produced by after mutation.eEven if the best individual is not picked to be a parent, it may still occur in the set of offspring.??Correct. Survival of the fittest is a very much over-used term to describe the largely random processes of evolution. Itonly definitely happens in Evolutionary Algorithms if you specifically choose to incorporate it. uHolland’s Simple Genetic Algorithm uses Fitness Proportional selection (with replacement) to pick n parents from a population of size n. From these parents n offspring are produced by one-point mutation, and bitwise mutation to become the next generation. Which of these statements are true about survival of the fittest in this algorithm?Which of the followi 9ESorry try again. Hint: there are two false possibilities in the list.111111.000000.111000.110011.011110.001111.?? Well done.oWhich of the following offspring can not be created by one point crossover from two parents 000000 and 111111 ?Which of the followi10ssorry. Rememebr that many different crossover operators have been designed to mix the genes present in two parents.111111000000111000.110011??Ycorrect - crossover cannot re-introduce gene values that were not present in the parents.eWhich of the following offspring can not be created by crossover from two parents 000000 and 001111 ?A population of indiSorry, try again.It stays constant.It slowly drops to 0%.It slowly rises to 100%.'It rises exponentially quickly to 100%.Correct. Crossover cannot change the value of single genes, thatis one role of mutation. Neither can crossover change their frequency atwhich individuals carrying those genes are selected to be parents, that is the role of selection. What crossover can do is to change the combinations of gene values that occur in different positions in an individual, but this is not relevant if we are onlty considering a single position.A population of individuals are all binary strings of length L. At time t, 30% of the genes in a certain position have thevalue 0, and 70% value 1. Those individuals with value 1 in that position are fitter than those with value 0. The population is now put in some random order, and one-point crossover is applied to consecutive pairs, i.e. matching strings 1 and 2, 3 and 4, etc. Two offspring are produced from each mating and these replace their parents. This process is repeated one hundred times, so it is equivalent to running a Genetic Algoritrhm for one hundred generations without parent selection or mutation. How does this frequency of 0s in the position of interest change as a result of this process?For which of the folFSorry, try again. Consider whether all of the offspring will be valid.kA permutation representing the order in which a series of operations are performed in an operating theatre.A binary string.LA sequence of integers representing moves from the set {left, right, ahead}.OA vector of floating-point numbers representing angles within a design problem.*Well done. All the types of representations except permuitationscan happily have the genes recombined by 2-X and still give valid offspring. In contrast, for example usinmg two-point crossover on the permutations12345 and 54321 will end up with duplicates of some values and none of others.mFor which of the following types of problem representation would it not be suitable to use 2-point crossover?A genetic algorithmSorry.1/27correct. The important point here is that for binary problems mutation is performed independently in each position. A random number is drawn and compared to the mutation rate. If it is below that rate the gene has its value changed.26/27.No. That would be probability that a randomly created number was higher than the mutation rate. Since p_m is low (1/27), what is most likely to happen in this case? Are we likely to be able to keep learned patterns?KIt is not possible to say without knowing what happened to the other genes.sNo. The important thing to remember is that for binary coded problems, each gene is mutated (or not) independently.1/17No. The gene position is irrelevant in this case. Mutation usually treats every gene in the same way, even if its effect may vary according to what random numbers are chosen. Well done.A genetic algorithm has individuals coded as binary strings of length 27. Mutation is applied with a bit-wise probability of 1/27. What is the probability that the gene 17 is changed by mutation?A genetic algorithm 2Sorry, try again. you need to think about the probability that any gene is not mutated, and about how to combine the probabilities of these independent events.1/L.0.1 - 1/L( 1 - 1/L ) L ;Correct. Note this quickly tends toward 0.5 as L increasesA genetic algorithm has individuals coded as binary strings of length L. Mutation is applied with a bit-wise probability of 1/L, i.e., 1.0 divided by L. What is the probability that a given chromosome will not be changed by mutation?For which of the fol 2TSorry, try again. Consider whether an offspring will be if just one gene is changed.kApermutation representing the order in which a series of operations are performed in an operating theatre.OA vector of floating-point numbers representing angles within a design problem.MA sequence of integers representing values from the set {left, right, ahead}.A binary string.~Well done - all of these types of representations can happily have just one gene value changed and still give valid offspring.For which of the following types of problem representation would it not be suitable to use a mutation operator which operated independently in each position?You are given the fiSorry, try again. Remember that probability of selection is proportional to fitness. Also, since exactly one a, b or c must get picked, all the probabilities must sum to one.14/(11+14+19).1/3.2/3.14 /(11+12+13). 14/(1+2+3).:Well done. Compare this to the same situation without the addition of 10 to the fitness. The effect of adding a constant value is to make all the selection probabilities much more similar. This in turn reduces selection pressure towards the fitter individuals, and makes it more likely they will be lost by chance.3You are given the fitness function f(x) = x2, +10 and a population of three individuals {a,b,c}. When decoded their genes when decoded give the values 1, 2 and 3 respectively. When you pick a single parent using Fitness Proportionate Selection, what is the probability that it is b?Is the followng statSorry, incorrect. Selection only works on fitnesses, which are a property of what is called the phenotype. The chosen representation, which is also known as the genotype, has to be decoded to give the solution (phenotype) before fitness evalaution can take place.correctIs the followng statement true or false? The action of the selection operator in an Evolutionary Algorithm is unrelated to the choice of problem representation.You are given the fi 2Sorry, try again. Remember that probability of selection is proportional to fitness, and that all the probabilities must sum to one, since one of {a,b,c} must get picked. 4/(1+4+9).Correct.2/3.^No. Remember that the probability assigned to each individual depedns on its relative fitness.1/3.^No. Remember that the probability assigned to each individual depends on its relative fitness. 4 /(1+2+3).^No. Remember that the probability assigned to each individual depends on its relative fitness.2 / (1 +4 +9).^No. Remember that the probability assigned to each individual depends on its relative fitness. Well done.IYou are given the fitness function f(x) = x 2, and a population of three individuals {a,b,c}. When decoded, their genes give the values x =1, 2 and 3 respectively. When you pick a single parent using Fitness Proportionate Selection, which of these expression correctly gives the probability that it is b?Doctor X has collect 2Sorry, try again.Using an Evolutionary Algorithm to evolve the structure of a multi-layer perceptron, with back-propagation to learn the weights.Taking a multi-layer perceptron with a pre-determined structure, and then using a real-coded Genetic Algorithm to evolve the weights.}Using genetic programming to produce a tree-based classifier, with measurement variables as some of the teminal (leaf) nodes.DUsing particle swarm optimisation to evolve a tree-based classifier.9Using an Artificial Immune System to create a classifier.?????YWell done. All of these methods could be used, probably with varying degress of success.Doctor X has collected a database of measurements of skin lesions such as size, colour, circumference, measure of smoothness, etc. The database also contains the the biopsy result (malignant or benign) for each lesion. You are asked to produce a diagnostic assistance system. When given the measurements from a newly presented skin lesion, this should predict whether it is malignant or benign. Which of these methods could typically be applied to produce the output for a given set of measurements?A number of on/off sSorry, try again.A string of n binary values.%A vector of n floating point numbers.:A string of values each coming from the set {1,…,n}.&A tree with n terminal nodes (leaves).-A permutation of the numbers {1, …, n}.OWell done. A typical problem for solving with a binary coded Genetic algorithm.SA number of on/off switches control a nuclear power plant, and a given configuration can be thought of as a state. It is desired to search the space of possible states to find one that minimises temperature fluctuations within the plant. What representation do you think would be most suitable for this problem if there are n switches?A mountain bike desiRSorry, try again. Are you unnecessarily restricting the set of possible solutions?)A vector of (n+m) floating point numbers. A string of (n+m) binary values.+A string of n values each between 1 and m.*A tree with (n+m) terminal nodes (leaves).0A permutation of the numbers 1 through to (n+m).gWell done. A typical problem for solving with a real-coded Genetic Algorithm or an Evolution Strategy. A mountain bike designer is trying to create a frame with certain desirable characteristics under simulation. To do this they must specify a set of n tube lengths and m angles between them. What representation do you think would be most suitable for this problem?It is necessary touSorry, try again. Recall that each patient will be scheduled exactly once, and your algorithm has to deal with this.$A permutation of the numbers 1 to n.A string of n Binary values.%A vector of n floating point numbers.4A string of values, each coming from the set 1 to n.&A tree with n terminal nodes (leaves).UWell done. This is a classic permutation problem of the “ordering” type.It is necessary to schedule a set of appointments with a doctor so as to minimise the average waiting time per patient. There are n patients. What representation do you think would be most suitable for this problem?A graph colouring ta]Sorry, try again. The best solution will naturally assign exactly one colour to each country.6A string of c values, each coming from the set 1 to n.&A string of (n times c) binary values./A vector of (n times c) floating point numbers.;A tree with n terminal nodes (leaves) and c internal nodes.$A permutation of the numbers 1 to n.Well done. Some other solutions might have worked e.g. binary and possibly the vector. However these would have been a lot less efficient and would have created extra artificial constraints.'A graph colouring task consists of assigning one of a fixed set of colours to each of the countries in a map, so that no two adjoining countries have the same colour. Assume that there are n colours and c countries. What representation do you think would be most suitable for this problem?6Fill in the blanks.Tbad luck
triple,statementsubjectobjectpredicate,propertywell done.
BFill in the blanks.

The basic component of knowledge representation in RDF is the [triple,statement]. This consists of 3 elements, the [subject], which is the resource being talked about, the [object], which is the last item in the statement, and the [predicate,property], which links the two.
0>The basic componentsorrytriplesubjectobjectproperty well done.The basic component of knowledge representation in RDF is the [triple]. This consists of 3 elements, the [subject], which is the resource being talked about, the [object], which is the last item in the statement, and the [property], which links the two.Give three examples0Correct examples might include:

1. The sky is blue eg is(sky,blue) in prolog notation; OR

2. Jim is Bob's brother eg brother_of(jim,bob) in prolog notation; OR

3. 'brother' is a symmetric property eg :brother a SymmetircProperty in Turtle (semantic web) notation;

NGive three examples of typical facts that might be stored in a knowledge base.0The criteria for a Ksorry. expressive effective efficientexplicit well done.The criteria for a Knowledge Representation Language (KRL) include: [expressive] – You can say what you want to say. [effective] – can infer things from what you have written down. [efficient] - reasoning should be possible in reasonable time and computational power. [explicit] – The KRL should be natural and easy to understand and if you have just inferred something, you should be able to answer the question ‘why’.6The criteria for a K 2&Remember all 4 answers start with 'e'.expressive,Expressiveeffective,Effectiveefficient,EfficientExplicit,explicit well done.

The criteria for a Knowledge Representation Language (KRL) include:

[expressive,Expressive] – You can say what you want to say.

[effective,Effective] – can infer things from what you have written down.

[efficient,Efficient] - reasoning should be possible in reasonable time and computational power.

[Explicit,explicit] – The KRL should be natural and easy to understand and if you have just inferred something, you should be able to answer the question ‘why’.

6Some sentences can hd

Remember: syntax is what it looks like (or how it is structured): semantics is what it means.

syntaxsemantics,semanticfwell done. Can you think of other ambiguous sentences? Where might this cause a problem for a machine?Some sentences can have the same [syntax] but quite different [semantics,semantic]. An example would be the sentence "They can fish" which could refer to the ability of fisherman, or to the activity of people in a food factory.Is this a sound infeBad luck - try again.
=well done. This is unsound reasoning, following from 'A implies B' to 'if B is true then A must be true" - technically called abduction. Although unsound, such techniques do have their uses in making guesses about what might be the cause of a given symptom.
Is this a sound inference?
  1. socrates is mortal;
  2. all men are mortal;
  3. therefore socrates is a man.

6A transitive propertjNo. A transitive property is something like 'ancestor' - the ancestor of my ancetsor is my ancestor.
ACwell done.
ZA transitive property is a property P such that if P(A,B) and P(B,C) then P([A],[C])
6A [symmetric,Symmetr&These are important definitions.
symmetric,Symmetrictransitive,TransitiveReflexive,reflexive1well done. These are important definitions.
A [symmetric,Symmetric] property is one in which P(A,B) implies P(B,A).
A [transitive,Transitive] property is one in which P(A,B) and P(B,C) implies P(A,C).
A [Reflexive,reflexive] property is one which applies to oneself hence P(A,A).
Match up these eleme|These are important technical terms in knowledge representation. Make sure you understand the concepts as well as the terms.Logical connective =>.Disjunction V.Conjunction ^.Universal quantifier.Existential quantifier. Negation.&Also known as NOT and written ! or .9Implies; so if A implies B and A is true, then B is true.mFor all ; as in; for all X if X is a cat then X is a mammal; To put it more succinctly "All cats are mammals.There exists - as in; if X is a parent then there exists a Y such that Y is a child of X. Or to put it more succinctly; parent has (at least one) child.8logical AND ; Leo is a lion AND a male; (must be both) .7logical OR ; Leo is staff OR student; (could be both) . well done.8Match up these elements of logic with their definitions.0A [symmetric] properSorry. debatable symmetric transitive reflexive Well done.A [symmetric] property is one in which P(A,B) implies P(B,A). A [transitive] property is one in which P(A,B) and P(B,C) implies P(A,C). A [reflexive] property is one which applies to oneself hence P(A,A).6In semantic nets, thobad luck. Though semantic nets are old, the isa and ako links are part of pretty much all modern KR techniques.isaakokind ofBwell done. semantic nets are an important early KRL to know about.

In semantic nets, there are two key relationships, or links:

[isa] which is used between instances and classes .

[ako] which is used between classes and superclasses and means 'a [kind of]' .

0In semantic nets, th 2obad luck. Though semantic nets are old, the isa and ako links are part of pretty much all modern KR techniques.classesisa instancesclassesakoclasses superclasseskindofKFeedback well done. semantic nets are an important early KRL to know about.In semantic nets, there are two key relationships, or links: [isa] which is used between [instances] and [classes]. [ako] which is used between [classes] and [superclasses] and means 'a [kindof]' .Match up the KR statSome of these definitions are a little advanced. Try to remember at least the basic ones (facts, rules, ontologies, subclass, domain and range).The sky is blue.XIF the blood sugar levels are high AND insulin levels are low THEN patient has Diabetes. A Student is (a type of) Person."Height is measured in centimetres.Find all my descendants.+Your email address uniquely identifies you.6The sun will rise tomorrow because it rises every day.>A formal description of the metaknowledge in a knowledge base.Fact.Rule. Subclass.Range.Transitive Closure.Inverse Functional Property. Induction. Ontology. well done!9

Match up the KR statements with their description.

6Knowledge Representa

These are 3 important modern techniques. Please let me know if you feel you got them right but under a different spelling...

Gobject orientation,unified modelling language,unified modeling language relationalsemanticgood.Knowledge Representation techniques have become key elements in modern computing approaches such as [object orientation,unified modelling language,unified modeling language] (UML), [relational] databases and the [semantic] web.6Fill in the blanks tgbad luck. Some of the techniques are quite advanced, but all are quite heavily used in current systems.SubclassDefault Functional functional SubpropertyDomainRange Transitive{well done! Would you be able to do it the other way round? (given a technique, come up with a good definition or example?).Fill in the blanks to complete this sample of techniques used in ontology reasoning:

[Subclass] reasoning – count the number of people in the room. Include all students, because they are people too.

[Default] reasoning: Birds fly. A sparrow is a bird therefore a sparrow flies. (you can also have default with override to allow for eg ostrich and penguin).

[Functional] reasoning: You can only have 1 spouse, I know that Jim is married to Thelma; on facebook he is listed as being married to xanga. I assume that xanga is the screen name of Thelma.

[functional] reasoning. I have found a facebook account with the email address jim@jimbob.com. I also find a LiveJournal account with the same email address. I surmise they describe the same person.

[Subproperty] reasoning. If I like Jim, it follows that I know Jim.

[Domain] reasoning. Only people can get married, Leaf is married to Tarquemelda therefore Leaf is a person. (NB not necessarily Tarquemelda).

[Range] reasoning. You can only get married to a person. Leaf is married to Tarquemelda therefore Tarquemelda is a person. (NB not necessarily Leaf).

[Transitive] closure. Find all my descendants.6MFill in the blanks t 2@Bad luck. Notice Induction is also known as Inductive reasoning. Deductive Inductive Abduction good stuff.Fill in the blanks to complete this set of definitions:

[Deductive] reasoning – If the initial facts are correct then the conclusions are sound.

[Inductive] reasoning– Reasoning based on past experience eg the sun will rise tomorrow (or strictly will appear to rise) because it rose yesterday, and the day before ...

[Abduction]. Reasoning along the following lines: All men are mortal. Socrates is mortal therefore maybe socrates is a man?0Fill in the blanks t 3gbad luck. Some of the techniques are quite advanced, but all are quite heavily used in current systems. heuristicSubclassDefault Functionalifp SubpropertyDomainRange Transitive{well done! Would you be able to do it the other way round? (given a technique, come up with a good definition or example?).Fill in the blanks to complete this sample of techniques used in ontology reasoning:

[Subclass] reasoning – count the number of people in the room. Include all students, because they are people too.

[Default] reasoning: Birds fly. A sparrow is a bird therefore a sparrow flies. (you can also have default with override to allow for eg ostrich and penguin).

[Functional] reasoning: You can only have 1 spouse, I know that Jim is married to Thelma; on facebook he is listed as being married to xanga; I assume that xanga; is the screen name of Thelma.

[ifp] reasoning. I have found a facebook account with the email address jim@jimbob.com. I also find a LiveJournal account with the same email address. I surmise they describe the same person.

[Subproperty] reasoning. If I like Jim, it follows that I know Jim.

[Domain] reasoning. Only people can get married, Leaf is married to Tarquemelda therefore Leaf is a person. (NB not necessarily Tarquemelda).

[Range] reasoning. You can only get married to a person. Leaf is married to Tarquemelda therefore Tarquemelda is a person. (NB not necessarily Leaf).

[Transitive] closure. Find all my descendants.Is this a sound infe 2This is perfectly sound reasoning, given the premises. The fact that the premise 2 sounds nonsensical does not mean the argument is unsound.Well done. This is perfectly sound reasoning, given the premises. The fact that the premise 2 sounds nonsensical does not mean the argument is unsound.

Is this a sound inference?

  1. socrates is a man;
  2. all men are teapots;
  3. therefore socrates is a teapot.
First order logic isCFOL is sound, complete and semi decideable. See Coppin p200.Sound; Complete; Decideable.??-well done. Do you know what these terms mean?First order logic is:0Here are some key deThese are complex ideas and at this stage you are only expected to grasp the basic intuitions. Make sure you understand them, refer to eg Coppin p200.soundtruetrue decideablesemidecideablecompleteinfertrue)Well done. These are quite complex ideas.Here are some key definitions in logic. Fill in the blanks to complete the definitions: (FOL = first order logic):

A logic is [sound] if, given true premises, everything we infer is true. (This property is [true] for both FOL and propositional logic).

A decideable logic allows us to ask the question: "Is it [true]?". Propositional logic is [decideable]. FOL is [semidecideable] which means that statements we cannot prove as true or false will take forever to ‘decide’.

A logic is [complete] if, informally speaking, we can [infer] everything that is true. This property is [true] for both FOL and propositional logic.6Here are some key de 2These are complex ideas and at this stage you are only expected to grasp the basic intuitions. Make sure you understand them, refer to eg Coppin p200.soundtruetrue decideablesemidecideable,semi-decideablecompleteinfertrue well done!Here are some key definitions in logic. Fill in the blanks to complete the definitions: (FOL = first order logic):
A logic is [sound] if, given true premises, everything we infer is true. (This property is [true] for both FOL and propositional logic).
A decideable logic allows us to ask the question: "Is it [true]?". Propositional logic is [decideable]. FOL is [semidecideable,semi-decideable] which means that statements we cannot prove as true or false will take forever to ‘decide’.
A logic is [complete] if, informally speaking, we can [infer] everything that is true. This property is [true] for both FOL and propositional logic.6^In a semantic net, w,These are the key concepts in semantic nets.nodesarcsisaako=well done. These are the key concepts in semantic nets.
In a semantic net, we create a network out of objects, represented as [nodes], and properties, represented as [arcs]. We can thus represent statements such as: Sammy [isa] sparrow; and Sparrow [ako] bird.0In a semantic net, w 2NSorry. These are the key concepts in semantic nets so it is good to know them.states transitionsnodesarcsisaako7Well done. These are the key concepts in semantic nets.In a semantic net, we create a network out of objects, represented as [nodes], and properties, represented as [arcs]. We can thus represent statements such as: Sammy [isa] sparrow; and Sparrow [ako] birdMatch each form of kThis should be straightforward.EA simple way of representing knowledge in the form of nodes and arcs.bA form of knowledge representation in which instances have slots, defaults and daemaon procedures.4The canonical representation for symbolic knowledge.3Representation of knowledge in the form of triples.semantic nets.frames.logic. semantic web.*well done. This should be straightforward.@Match each form of knowledge representation with its definition.Some problems with spMake sure you understand what semantic nets can and cannot do; and also what these types of reasoning are.
)The properties have weak semantics.
&The properties have weak syntax.
'They do not do default reasoning.
9Their default reasoning does not handle exceptions.
(They cannot do subclass reasoning.
6They do not cleanly handle multiple inheritance.
???well done.
/Some problems with semantic nets include:
What are some of theGBad luck. This question covers most of the key concepts in frames
,Default values which can be overridden
"Classes, instances and slots
Clearly defined properties
&A mechanism to handle exceptions
1A mechanism to support procedural reasoning
.A solution to the class/instance problem
????Qwell done! You should now be able to write a short description of frames.
2What are some of the capcbilities of frames?
What sorts of reasonbThis is testing your knowledge of reasoning techniques as well as semantic nets and frames.
Subproperty reasoning
Subclass reasoning
Default Reasoning
inverse property reasoning
Cardinality
Existential Quantifiers
Universal Quantifiers
??well done.
HWhat sorts of reasoning are supported by semantic nets and frames?
6Frame modelling is tAbad luck. It's good to know how frame logic relates to UML.
UMLUnified Modelling Languageinstanceclassslotswell done.
7Frame modelling is the precursor to what is known today by a 3 letter acronym [UML] which stands for [Unified Modelling Language]. In this technique we have the notion of objects each of which is an [instance] of a [class]. Modern day object attributes correspond directly to the frame notion of [slots].
0Frame modelling is t 2\It is good to understand how object orientation stems from ideas in knowlege representation.HTMLUniversal Markup LanguageHypertext Markup LanguageclassslotsmethodsUMLUnified Modelling LanguageinstanceexamplecategorygWell done. It is good to understand how object orientation stems from ideas in knowlege representation..Frame modelling is the precursor to what is known today by the acronym [UML] which stands for [Unified Modelling Language]. In this technique we have the notion of objects each of which is an [instance] of a [example]. Modern day object attributes correspond directly to the frame notion of [category].An expert system conIYou might like to carefully work out a 'family tree' to help you.
ancestor(sue,dave).ancestor(john,dave).
ancestor(dave,john).
ancestor(charles,dave).
ancestor(alice,dave).ancestor(bob,dave).ancestor(sue,john).
ancestor(charles,john).??????well done.
An expert system contains the following facts and rules.

1. IF ancestor(X,Y) AND ancestor (Y,Z) THEN ancestor (X,Z).
2. IF parent(X,Y) THEN ancestor(X,Y).
3. IF son(X,Y) THEN parent (Y,X).
4. son(john,dave).
5. parent(sue,dave)
6. parent(alice,john).
7. parent(charles,sue).
8. parent(bob,alice).

Which of the following facts are true?6An expert system con 2ZBad luck. This question tests terminology - do you understand the underlying ideas?
factsrulesruleantecedent,precedent consequentTwell done. This question tests terminology - do you understand the underlying ideas?An expert system contains a series of [facts] which are statements about the world, and [rules] which act on these statements.
The left hand side of a [rule] is called its [antecedent,precedent] and the right hand side is called the [consequent].
0sAn expert system con 3"You need to know this terminology.factsrules antecedent consequent-Well done. You need to know this terminology.An expert system contains a series of [facts] which are statements about the world, and [rules] which act on these statements. The left hand side of a rule is called its [antecedent] and the right hand side is called the [consequent].An expert system con 4Note rule 3 which might have been more intuitive as IF son(X,Y) THEN parent(Y,X). Of course then the answers would have been different!
ancestor(sue,dave)ancestor(john,dave)ancestor(dave,john)ancestor(charles,dave)ancestor(alice,dave)ancestor(bob,dave)
ancestor(sue,john)ancestor(charles,john)?????!well done. This was tricky!
An expert system contains the following facts and rules.

1. IF ancestor(X,Y) AND ancestor (Y,Z) THEN ancestor (X,Z).
2. IF parent(X,Y) THEN ancestor(X,Y).
3. IF son(X,Y) THEN parent (X,Y).
4. son(john,dave).
5. parent(sue,dave)
6. parent(alice,john).
7. parent(charles,sue).
8. parent(bob,alice).

Which of the following facts are true?
6This question dealsTMake sure you understand the concepts here, even if you chose different words.
firstrandomspecificrecentlycontextwell done.
lThis question deals with conflict resolution, when multiple rules match a given condition. How should we choose which rules fire? (or at least fire first). Fill in the blanks.

The easiest approach is just choosing the [first] rule to match. This is simple and deterministic. An equally simple, but stochastic method is to choose one rule at [random]. Sometimes we choose the rule which is the least general - ie most [specific]. Or we can take the history into account, taking the most (or least) [recently] used rule. Finally we can use the notion of [context] - a group of rules related to the task at hand.0This question deals 2sorry.firstrandomspecificrecentlycontext Well done.dThis question deals with conflict resolution, when multiple rules match a given condition. How should we choose which rules fire? (or at least fire first). Fill in the blanks. The easiest approach is just choosing the [first] rule to match. This is simple and deterministic. An equally simple, but stochastic method is to choose one rule at [random]. Sometimes we choose the rule which is the least general - ie most [specific]. Or we can take the history into account, taking the most (or least) [recently] used rule. Finally we can use the notion of [context] - a group of rules related to the task at hand.Match up the followi5These are some key concepts for expert systems.
Rule
Fact MetaknowledgeForward chaining
Backward chaining
Antecedent#The left hand side of a rule.
CAn expert system components that links facts to consequents.
Data driven reasoning.
Query driven reasoning.
WKnowledge about the rules and facts - for example methods of conflict resolution.
An assertion.
well done.
HMatch up the following expert system terms with their definitions.
6There are many typesmThis is basically a summary of the KR subtopic so make sure you know where the various languages stand.
semantic netframesexpert systems semantic webresource description frameworklogicfull house! well done.
There are many types of knowledge representation languages. For example, a simple method, introduced in the 1960s used a graph based representation. It was called a [semantic net]. In the 1970s, [frames] were introduced and laid the theoretical foundations for modern day techniques. [expert systems] use rules for their reasoning. The [semantic web] uses a language called RDF which stands for [resource description framework]. All the above methods can equivalently be represented in some form of [logic].
Here are some statemSsome of these were tricky. Note expert systems *are* a logic based technique.
PForward chaining typically yields fewer new facts than backwards chaining.
BExpert systems are an alternative to logic based techniques.
AConflict resolution strategies are a type of metaknowledge.
5Expert systems can be used for fault diagnosis.
??well done.
DHere are some statements about expert systems, Which are true?
00There are many types 2gThis is basically a summary of the KR subtopic so make sure you know where the various languages stand.finite state machineobjectsTuring MachinesNeural networksrelational database formatinternet semantic netframesExpert systems semantic webresource description frameworklogic Well done.There are many types of knowledge representation languages. For example, a simple method, introduced in the 1960s used a graph based representation. It was called a [semantic net]. In the 1970s, [frames] were introduced and laid the theoretical foundations for modern day techniques. [Expert systems] use rules for their reasoning. The [semantic web] uses a language called RDF which stands for [resource description framework]. All the above methods can equivalently be represented in some form of [logic].6Fill in the blanks i 3+Try checking out the W3C pages on RDF
data,information,knowledgew3c,worldwide web consortiumresource description framework well doneFill in the blanks in this definition:

The Semantic Web provides a common framework that allows [data,information,knowledge] to be shared and reused across application, enterprise, and community boundaries. It is a collaborative effort led by the [w3c,worldwide web consortium] with participation from a large number of researchers and industrial partners. It is based on the [resource description framework] (RDF).

Hint: You can easily find this definition on the web!

6Fill in the blanks:Itriplesubjectobjectpredicate,propertyDFill in the blanks:

In RDF, the basic unit of knowledge representation is a 3 element statement, or [triple]. This statement comprises a [subject], the resource being talked about, an [object], which generally appears at the end of the statement, and a [predicate,property] which linkes the two.
Which of the followi11(These are the key characteristics of RDF>In RDF, knowledge is represented in the form of triples.
>RDF can be represented as a directed acyclic graph (DAG)
0RDF can be represented as a directed graph
.RDF can be written down in an XML format
-RDF can only be written down in an XML formatbResources, except anaonymous resources, are identified by globally unique identifiers (URIs)
????9well done, these are the key characteristics of RDF
3Which of the following statements is correct?
Here are some proper>These are some more advanced details of RDF.
?RDF can be written down as a set of triples or as a graph
+The order of triples is not important
An RDF graph is directed
An RDF graph is typed
An RDF graph is acyclic
+Subjects can be resources or literals
*Objects can be resources or literals
-Properties can be resources or literals
BAn object of one triple can by the subject of another triple
DThe subject of one triple can be the subject of another triple
EThe property of one triple can be the subject of another triple
????????8Good. These are some more advanced details of RDF.
6Here are some properties of RDF. Which are true?
0Complete the followi 2^This is a summary of what you are expected to know about the semantic web at this stage.
information modellingdata integration vocabulariesRDFtriplesgraphschemaweb ontology languagerangeproperty cardinalityuniform resource identifierwell done.
Complete the following paragraph:

The semantic web really has 2 main elements. 'Semantic' means [information modelling], a way of writing down rich and precise descriptions of knowledge. 'Web' means global [data integration], a framework for sharing both data and models (we often call these [vocabularies]). The semantic web language, [RDF], allows us to represent knowledge in the form of [triples] which can be connected together into a directed, typed [graph]. The semantic web ontology languages RDFS (RDF [schema]) and OWL ([web ontology language]) allow us to specify increasingly rich semantics such as domain and [range], subclass and sub[property], [cardinality] (number of values for a property) and so on. Resources are assigned a globally unique identifier called a URI ([uniform resource identifier]) which means that data from different sources can be merged without fear of ambiguity.
Which of these tasksjAll these can be tackled by neural networks except for #4 and #6. Try again and you'll find out why.
FChecking credit card applications for possible examples of fraud
(Recognising a face in CCTV footage
#Filtering your email for spam
MExplaining why a set of medical tests suggests the patient has diabetes
%Controlling a robot's movements
1Finding the optimal route between 2 cities.
????$Good. All these can be tackled by neural networks except for: number 4, which requires explicit representation of reasoning (eg logic); and number 6 which is a search/optimisation based approach (one might use A* for a guaranteed optimum or genetic algorithms for a heuristic approach).
BWhich of these tasks might be suitable for a neural network?
6#A real biological nesoma dendritesaxonsynapseweightwell done
A real biological neurone comprise a cell body, or [soma], which tentacle-like inputs (called [dendrites]) and one long output ([axon]). 2 neurones are joined together by a junction called a [synapse]; this is where the output of one neurone is passed to the input of another. The efficacy of this junction depends on many factors including anatomy, electrical and chemical properties and other neuronal activity - thus we say that each neuronal connection has a [weight].
0uNeural networks are?This is the definition of a neural network in a nutshell.
 subsymbolicsymbolicexplicitobjects relationshipssignalsoutputwell done
Neural networks are an example of [subsymbolic] processing - as opposed to the knowledge representation techniques we covered last term, which are [symbolic]. That means that they do not use [explicit] identifiers, like 'bob', 'brother_of' and 'X' to refer to real world [objects] and [relationships]. Instead [signals] from the environment are sensed, weighted, combined and compared to a threshold in order to produce an [output].
Show how some biologbad luck.
DendritesAxons%Strength of synaptic connection
Neurone
Learning
Action Potential fired
 Input
Output weight
unitlearningOutput = ON.
good.fShow how some biological characteristics of real brains are mapped to artificial neural networks
Put these quantitiesXNo. The answers range from less than one thousand to well over a thousand billion.
<Number of units in a typical commercial neural network
(Age of the earth (in years)

.Number of nerve cells in the human brain
+Number of synapses in the human brain
3Yes. the answers range from <1000 to 10^13
<Put these quantities in order (from lowest to highest)
Match up theseneuraXYou are allowed to use google! Remember that one answer needs to be used multiple times.McCulloch and Pitts Rosenblatt Rumelhart, McClelland and Hinton Bryson and HoMinsky and Papert Paul WerbosBilliard ball neurone modelThe perceptronBackpropagation['Perceptrons' a book showing the workings of, and crucially the limitations of, perceptronsxYes. Note that backprop was 'invented' several times; Bryson and Ho in 1969, Werbos in 1974 and Rumelhart et al in 1986.Match up theseneural network pioneers withtheir chief invention. You are allowed to search the web! You'll have to use one answer more than once.0In the 'billiard bal_Remember this is the 1st model we talked about, the 'logical calculus' of McCulloch & Pittsbinary+10+1-10bias+1 well done.In the 'billiard ball' model, all inputs are [binary], that is they can be on [+1] or off [0]. Weights can only have two values, [+1] and [-1]. The threshold of a neurone is strictly speaking always [0] but can be effectively changed using [bias] inputs which are inputs clamped to [+1].Which of the followi12 bad luck.LThe billiard ball model (logical calculus) can compute any logical function.0The perceptron can compute any logical function.9A multilayer perceptron can compute any logical function.NIt is necessary to compute any logical function to tackle real world problems.??? well done."Which of the following is correct?04The perceptron train'There are some 'red herrings' in there!biasoutput thresholdweightfixedtargetactualbinaryinputerror well done

The perceptron training law is simple. It takes the form:

change in[weight] = learning rateMULTIPLIED BYerrorMULTIPLIED BYinput.

where the learning rate is [fixed]; the error is the [target] output minus the [actual] output; and the input is often [binary]. This means that there are two cases where no change in weight is applied: where the [input] is zero, or where there is no[error] (ie perfect response).

0/Complete the followi 3esorry. Try again to get this right - this sums up the basic ideas behind Evolutionary Computation.dinfinitedecreaseevolvespeciesniches population individualstraitsfinite competition resources hereditableincrease generationsVWell done. This sums up the basic notions driving biological and artifical evolution.Complete the following sentences by making appropriate selections from the choices presented. The central metaphor of evolution is that [species] adapt to fill environmental [niches]. A [population] of a given species will consist of many [individuals] with different characteristics. Some of these [traits] affect how well individuals are suited to exist in their environmental niche. In practice, environmental niches have a [finite] size, so this causes [competition] for [resources]. If the differences that cause the differences are [hereditable] (i.e. they have a genetic basis), and make individuals better able to compete for resources, then the number of individuals those traits will tend to [increase] in future [generations].