1 / 33

Chapter 10 Artificial Intelligence

10.1 Intelligence and Machines. Today's computers are capable of performing defined tasks with speed and accuracy.But, when faced with a situation not forseen by the programmer, a computer's performance deteriorates rapidly.In contrast, the human mind often flounders on complex computations but is capable of understanding and reasoning..

adamdaniel
Télécharger la présentation

Chapter 10 Artificial Intelligence

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


    1. Chapter 10 Artificial Intelligence By Mr. G. Flores And COSC 1309 Class Spring 04

    2. 10.1 Intelligence and Machines Todays computers are capable of performing defined tasks with speed and accuracy. But, when faced with a situation not forseen by the programmer, a computers performance deteriorates rapidly. In contrast, the human mind often flounders on complex computations but is capable of understanding and reasoning.

    3. Performance verses Simulation If we build machines that are able to perform complex tasks without human intervention, we must have them more humanlike in the sense that they must process (or least simulate) the ability to reason. Trying to build machines that reason has made Computer Scientists turn to branches of study like psychology and biology To find flexible principles to apply to programs

    4. Performance vs Simulation Consider a poker playing program that needs to be developed. CS would: Include stats & probability, bluff at random, show no emotion and try to maximize chances of winning Psychologist would: Include theories of human thought and behavior the result might be that it may play aggressively or be intimidated and thus involve emotion

    5. Performance vs Simulation Therfore: A Computer Scientist main concern while developing the program would be the programs Final Performance or Performance Oriented A Psychologist would be interested in the processes of natural intelligence and so would approach the project as an opportunity to test theories. From this point of view, the development of the intelligent program is actually a side effect of another pursuit progress in understanding human thought and behavior. This approach is said to be simulation oriented.

    6. The Turing Test If asked to decide whether the poker-playing programs developed by the computer scientist and the psychologist actually possess intelligence and if so, which program is more intelligent. Also, is intelligence measured by the ability to win or the ability to be humanlike?

    7. The Turing Test Alan Turing in 1950 adopted the idea of measuring if a program had the ability to be humanlike. The Turing Test: Goal: To evaluate the intelligent behavior of a machine. SetUp: allow a human (interrogator) to interact with a test subject by means of a typewriter system, without being told whether the test subject was a human or a machine.

    8. The Turing Test In this environment, a machine would be declared to behave intelligently in the event that the interrogator was not able to distinguish it from a human. Turing proposed that by the year 2000 machines would have a 30 percent chance of passing a five-minute Turing test a conjecture that has turned out to be surprisingly accurate.

    9. The Turing Test A well-known example of a Turing test scenario arose as a result of the program DOCTOR (a version of the more general system called ELIZA) developed by Joseph Weizenbaum in the mid-1960s

    10. The Eight Puzzle Machine Task is to investigate how machines can be programmed to appear to be intelligent Consider the design of a machine having elementary intelligence characteristics Box, gripper, camera and rubber end finger

    11. The Eight Puzzle Machine Puzzle (out of order) is placed on the machine Machine starts pushing around the squares in the puzzle until they are in order Machine releases the puzzle and turns itself off.

    12. 10. 2 Understanding Images Because such a machine involves elementary perception as well as reasoning abilities, its design provides a basis for presenting the topics of this section and others. The problem is not storing the image but actually to understand the image and extract the current status of the puzzle

    13. 10.2 Understanding Images Tasks of understanding general images is usually approached by a two step process 1 image processing 2 image analysis

    14. Production Systems A system of common characteristics to solving reasoning problems. Consists of three parts: A collection of states. A collection of productions. A control system.

    15. A collection of states State is a situation that might occur in the application environment. The beginning state is called the start (or initial) state. The desired state (or states) is called the goal state.

    16. A collection of productions A production is an operation that can be performed in the application environment to move from one state to another. Each production maybe associated with preconditions. Ex. If/Else and While Loops

    17. A control system The control system consists of the logic that solves the problem of moving from the start state to the goal state. The control system must decide which of those productions whose preconditions are satisfied should be applied next.

    18. Behavior-Based Intelligence Early work in AI approached the subject in the context of writing programs to simulate intelligence. Human intelligence is based on the evolution of behavior, therefore, human intelligence is better simulated by simple stimuli response functions.

    19. Production Systems State Graph A convenient way of representing or at least conceptualizing, all the states, productions, and preconditions in a production system. Uses a collection of nodes connected by arrows. Has a start state and a goal.

    20. Production Systems

    21. Search Trees

    22. Definition The root node of the search tree is the start state and the children of each node are those states reachable from the parent by applying one production. Each arc between nodes in a search tree represents the application of a single production and each path from the root to leaf represents a path between the corresponding state graph.

    23. Example of unsolved puzzle to be solved by a search tree:

    24. Sample search tree

    25. Solution of puzzle

    26. Why Need It? Using search trees becomes cumbersome with complex problems Developing an entire search tree becomes impractical Another strategy is to build the tree in a breadth-first manner Building the tree layer by layer as the program executes Results in depth-first construction

    27. Depth-First Pursuing the most promising path and then back tracking if first choice turns out to be a false lead Thus tree is constructed with vertical paths vs. horizontal We humans rarely pursue several options at once (breadth-first) Probably would select the option that appeared most promising and follow it We rarely know which option is best at a particular point Thus, we need a way of identifying which of several states is most promising

    28. What is Heuristic? A quantitative value associated with each state that attempts to measure distance from that state to the nearest goal Measure of projected cost Given a choice between two states, the one with the smaller heuristic can real its goal with least cost This state would thus represent the direction we should pursue

    29. The Two Characteristics It should have a reasonable estimate of the amount of work remaining in the solution if the next state is reached The better the estimate provided by the heuristic, the better will be the decisions that are based on the information It should be easy to compute Benefits the search process rather than being a burden In contrast, although the actual number of moves required to reach the goal from the given state is an excellent thing to know when making decisions, computing this information requires finding the actual solution first

    30. Comparing Distances A simple heuristic in the case of the eight-puzzle would be to estimate the distance to the goal by counting the number of tiles that are out of place A state in which four tiles are out of place is worse than having only two tiles out of place This heuristic does not take into account how far out of place the tiles are. If the two tiles are far from their proper places, many states will have to be passed through to get all the tiles in their correct positions

    31. Comparing Distances A better heuristic would then be to measure the distance each tile is from its destination and add these values to obtain a single number (quantity) A tile right next to its destination would have a distance of one A tile whose corner touches the space of its destination has a destination of two It must move one position horizontally, and one position vertically This heuristic is easy to compute and gives us a rough estimate of the number of moves required to reach the goal

    32. Decision-making Process Now to implement the heuristic Recall that a human faced with a decision tends to select the option that appears to be closest to the goal Our search should thus consider the heuristic of each leaf node, then chose the node with the smallest heuristic and proceed with the search from there The algorithm for this is to the next slide

    33. Algorithm for a Control System Using Heuristics Establish the start node of the state graph as the root of the search tree and record its heuristic value While (the goal has not been reached) do [Select the leftmost leaf node with the smallest heuristic value of all leaf nodes. To this node attach as children those nodes that can be reached by a single production. Record the heuristic value of each of these new nodes next to the node in the search tree ] Traverse the search tree from the goal node up to the root, pushing the production associated with each arc traversed onto a stack. Solve the original problem by executing the productions as they are popped off the stack.

    34. Applying the Algorithm Using the algorithm from the previous slide, we first establish this state as the root and record its heuristic value of 5 Then the first pass through the while structure gives us the 3 nodes that can be reached from the initial state Note that the heuristic value is placed below each of the nodes

    35. Applying the Algorithm The goal has not been reached, so we perform the while again, this time extending our search from the leftmost node

    36. Applying the Algorithm The heuristic value of the leftmost node is now five, indicating this may not be the best path Algorithm picks up on this and in the next pass the loop instructs us to expand the tree from the right most node The next leftmost node with the smallest heuristic value

    37. Applying the Algorithm At this point the algorithm seems to be on the right track and continues on that path as the heuristic value is 3, and then produces the tree to the right

    38. Applying the Algorithm Comparing the previous tree to the one on this slide, even with the wrong turn in the beginning, the heuristic approach greatly decreased the size of the search tree and produced a more efficient process

    39. Applying the Algorithm After reaching the goal state, the while terminates, and we move on to traverse the tree from the goal node up to the root, pushing the productions encountered onto the stack as we go Finally, we are instructed to execute these productions as they are popped from the stack At this point, we would observe the machine lower its finger and begin to move the tiles

    40. 10.4 Artificial Neural Networks

    41. Artificial Neural Networks were created because: Central processing units that execute sequences of instructions were not capable of perceiving and reasoning at levels comparable to those of the human mind. Many researchers turned to machines with other architectures (being ANN).

    42. Artificial Neural Networks are constructed by: They are constructed from many individual processors (called processing units). Processors work in a manner that models networks of neurons in living biological systems.

    43. Basic Properties A neuron in a living biological system. (pg.424) Dendrites a biological neuron is a single cell with input tentacles. Axonan output tentacle which decides whether the cell is in an inhibited or excited state. Synapsessmall gaps where dendrites pick up signals from the axons of other cells.

    44. A processing unit in an artificial neural network is a simple device that mimics this basic understanding of the biological neuron. Inputs are denoted by v1, v2, v3. Weights are denoted by w1,w2,w3. Formula is represented by (v1w1+v2w2+v3w3) If the total is greater than the value on the threshold value it fires(produces a 1), else it produces a 0. Examples: weights inputs:-2,3,-1 & output units:1,1,0 -> (1)(-2)+(1)(3)+(0)(-1)=1, thus its output is 0. Else, outputs are changed to 0,1,1(0)(-2)+(1)(3)+(1)(-1)=2, thus the output is 1 and it fires. Activities within a processing unit

    45. What is a genetic algorithm? The approach of this algorithm is to intermix the best performers within a collection of proposed solutions to obtain another generation of proposed solutions that represents improvements over the original collection.

    46. Evolutionary programming This is a learning program This is what is basically meant in the genetic algorithm technique That is why the functional paradigm is implemented

    47. Algorithm Evolution

More Related