IAI JOHN WOODWARD. PROBLEMS FOR DISCUSSION/THOUGHT Miscellaneous Given an example of a reflex response. (one example is the knee-jerk response) Given an example of a learnt response. (one example is ) Is your ability to program a good test of intelligence? A cork is pushed inside a wine bottle – how and you get it out? Given an example of a linear relationship (one that can be written as y=mx + c) and a non-linear relationship (one that cannot be written as y=mx + c for example y=a x^2 + b x +c). Please do not use an example from the lectures. Give two examples of feedback – and two different types (e.g. Boolean and continuous) e.g. shooting a target when you can/cannot see where the bullet hit (you only know the bullet hit) and parking a car – the helper bangs the car boot – or better shows you demonstrates the distance between the cars. Exams – you get a mark back (a percentage or pass/fail). Think of other things we “generate and test”? Examples include software, medicine, mathematical proofs, art, music, written text, speech (in our heads before we say it), photos (we discard the ones in which we do not look good). Artificial Neural Networks Briefly describe the difference between a perceptron, a 3-layer feed-forward Artificial Neural Networks, a ragged Artificial Neural Networks, and a recurrent Artificial Neural Networks. Artificial Neural Networks are different from conventional computing (e.g. C/Java and Turing machines), they are parallel, have gradual degradation properties and are sub-symbolic – what is meant by each of these? Which of the 16 2-input Boolean functions are linearly separable? You can do this graphically. Nature inspired computation. How do ants find their way back to the nest? Ant leave a pheromone trail – this is a smell other ants use to communicate. It is a language? “Smell” does not contain direction – so how can a pheromone trail help an ant find its way home? Artificial Neural Networks are not modelling the brain. Give an example of what we are trying to copy with ANN. What we are not trying to copy. For example, ANNS do not really model forgetting (which is a natural/normal part of human cognition). Genetic Algorithms are not modelling evolution. Give an example of something they are trying to copy from evolution and something they are not trying to copy. Visual Reasoning; Superman wears his pants over his trousers; therefore it is easy for him to remove only his pants. Real men wear their pants under their trousers. Can a man take off his pants without first removing his trousers? How did you arrive at your answer? Does this support the hypothesis that we think in English (or Chinese, or any other natural language). (For reference, the equivalent problem is can a man remove his vest (undershirt) without removing your shirt). Imagine the pants (vest) are very elastic. Is point X inside or outside? Problem Formulation Which of the following can we formulate as problems on a computer? In each case state briefly in a few sentences why or why not. Some of these problems are ill-defined. To formulate a problem we need to be able to do the following: Define a representation, Goal state, a set of Actions. a. b. c. d. e. f. g. h. i. Why are left and right reversed in a mirror – but not up or down. The spiral question? From above The European Debt Crisis. Chess. The North/South Korean problem. Voice recognition. Global warming. Why is a manhole/plughole round? Given nine points at the coordinates (i,j) where i,j are in the set {-1,0,1} connect all the dots with 4 lines. Learning by Common Sense In my bank account I could say I have $1234 USD (United States Dollars). This is reasonable and you may believe me. What about each of the following – explain your reasoning. j. k. l. m. n. o. 100.01 -345 123.4567 Root 3 9.0E999 (i.e. 9 followed by 999 zeros). Five