Fichas
Sin título
Fichas del conjunto (46)
- 1Search-based problem solverMakes decisions based on likely consequences of potential actions
- 2State SpaceAll legal problem states represented by a graph where states/nodes are connected by arcs representing legal moves
- 3Dynamic ProgrammingOnly need to remember best route to each state
- 4Combinatorial ExplosionEach state generates successor states from potential actions and their consequences, generating successors of successors quickly leads to a large increase in nodes causing a 'combinatorial explosion'
- 5Search Strategy RequirementsGUARANTEED to find a solution
ADMISSIBLE to find the best solution
EFFICIENT so solution nodes / total nodes - 6Depth-FirstRecursively expand on successor of previous iteration until goal reached or dead end, if dead end continue at last level with an alternative
- 7Breadth-FirstExplore each node at level I before moving to level I + 1, undirected but admissible if uniform costs
- 8Depth-Bounds Beam Search Branch-And-Bound Best-First A*Depth-First to level N
Breadth-First but best few each time
Breadth-First but accumulated cost from start
Estimated costs to goals
Accumulated costs + Estimated costs to goals - 9Hill ClimbingHas simulated annealing and genetic algorithm as alternatives, attempts to maximise objective function
- 10Dynamic Time WarpingA way of measuring similarity between an input speech pattern and stored templates of speech when they are of different lengths by calculating local distance measures
- 11Viterbi AlgorithmConnected word recognition problem, find best sequence of templates to match the data
- 12Constraint Satisfaction problemSet of variables with domains that are the values it can take, set of constraints to restrict the domains, solution is set of values to satisfy all constraints
- 13Skilled GameplayProblem solving abilities like strategy/planning, well defined, tractable mciro-world, let you test against people/machines, reproducable
- 14Requirements for gameplayWay of creating legal moves given board position, way of representing board positions, way of detecting terminal nodes, way to search/expand game tree
- 15Partial Look-Ahead / Minimaxing vs Back UpPartial Look-ahead relies on evaluating LEAVES and backing up the best result for each player
Back Up relies on evaluating TERMINAL NODES and backing up the best result for each player - 16Heuristic PruningMoves are ordered by plausibility, search effort expended according to ranking, lowest plausibility never considered
- 17Alpha-Beta pruningBack Up is performed but work is discontinued below nodes which wont change decision at level above
- 18Evaluation FunctionEstimates how good a position is for the next player.
Assumes independent features, static evaluation, trade-off between search depth and complexity, better for simpler positions, arguable to return board position into a number - 19Book MovesUse libraries of standard positions, problem is recognising when its not identical but similar
- 20Forward ChainingLook for facts that match antecedents of rule, deduce the consequents, look for more deductions
- 21Backward ChainingGiven hypothesis, attempt to find rule with consequents that match the hypothesis, then try prove its antecedents in the same way.
- 22Beyond The HorizonDelay but not prevent disasters, disaster occurs after depth you search to
Search past situations considered dynamic or grow a small secondary position around a considered move and compare the backed up value - 23Deduction Induction AbductionIf X then Y, have X deduce Y
Observe X1->Y X2->Y X3->Y Induce X->Y
Observe Y, hypothesize X - 24Within and Between cycle repetitionMatch rules that share antecedents duplicates work, so use rule network to capture this behaviour
Remember matches so you don't match again using tokens that store remaining antecedents and context of matched - 25Production SystemsImplement controlled forward chaining using short-term memory, if predicate fires and antecedents match, modifies context
- 26Conflict ResolutionIf there is more than one rule with antecedents and predicate satisfied, which fires?
- 27JAVALacks advantages of functional programming, treating functions as first-class objects
OOP is natural framework for storing knowledge, inheritance and slots are useful for AI
Does not support pattern matching in a useful way - 28PlanningYou still find sequence of actions to transform initial state to goal, but use knowledge about potential actions, how they change the state and their preconditions
- 29Means End AnalysisGet differences, find operator to reduce these, match preconditions of the operator in the same way, apply the operator
- 30STRIPSActList, what to add to plan if used, addList and delList of what is added/removed to current state, list of preconditions
3 STRIPS
STRIPS1 - Get differences
STRIPS2 - Get Operator
STRIPS3 - Apply Operator
Operators need to be in right order, goal conflict, no attempt to decide which goal to work on first - 31PerceptionReal intelligent systems have to function without a given knowledge base/problem in data structure
Make sense of data around them using learning and perception - 32Perceptual ProblemsAmbiguous Figures
Paradoxical Pictures
System Error
Distortion Illusions
Perceptual Invention - 33Auditory PerceptionTransform sound waves into signals, filter out background noise, turn sound patterns into recognisable percepts in the brain
- 34Cocktail Party Effect McGurk EffectMultiple sources, listen to one, filter rest
Audio/Visual cues clash, visual takes priority - 35Gestalt PrinciplesDescribe how humans perceive and group objects using proximity, similarity, continuance (eye prefers to follow continuous lines), closure, figure & ground (distinguish figure from background), etc
- 36Problem Solving using KnowledgeProblem given as data structure, same as knowledge
Problem solver applies symbolic manipulation on both to get a result - 37Computational Auditory Scene AnalysisGet input mixture, break into features, build up into discrete objects, group objects together, distinguish between groups
- 38Caregiver SpeechImproves learning, exaggerated information, slower tempo and longer pauses
- 39Top down/Bottom up processingInformation processing guided by X-level models like experience/expectations or loudness/frequency/etc
- 40Stream SegregationTwo cycles of sound with equally spaced notes, interleaving creates two perceptual streams with irregular rythms
- 41Why is perception importantPeople's behaviour is based on belief of reality not reality itself
We can develop AI further for speech recognition and visual perception - 42Gregory's MetaphorPerceptions explain data in arrangements of known objects and can be used to predict hidden parts of objects and/or the future
- 43Baseline SystemBase system you're improving on
- 44Independent/Dependent VariablesIndependent: One being controlled
Dependent: One that changes as a result of the independent variable changing - 45Significance TestingTests if data recorded is statistically significant
Done by calculating probability that the change in the dependent variable happened by chance or not - 46Sequence MatchingFind best sequence of models to match the data
Información
- Descripción
- Fichas sobre Sin título, creado por thomas jeffery el 25/05/2019.
- Creado por
- thomas jeffery
- Visitas
- 8
- Valoración