Pojmy: kartičky | uvod-do-umele-inteligence

introduction

symbolic AI
sylogism
utility
knowledge-based agent
control theory
cybernetics
microworld
perceptron
undirected research
expert systems
gartner hype cycle
rational agent
omniscient agent
environment
- fully observable
- partially observable
- deterministic
- stochastic
- episodic
- sequential
- static
- dynamic
- discrete
- continuous
- multi-agent
  - cooperative
  - cooperative

problem solving, search

agent
- rational
- reflex
  - simple
  - model-based
- goal-based
state representation
- atomic
- factored
- structured
problem solving agent
environment abstraction
- valid
- useful
well-defined problem
- initial state
- transition model
- goal test
search
- graph-search
- tree-search
- frontier
- uninformed
  - BFS
  - DFS
    - backtracking
  - uniform-cost search
- informed
  - best-first search
    - greedy
    - A*
- heuristic function
  - admissible (přípustná)
  - consistent/monotone
  - $h_1$ dominates $h_2$

constraint satisfaction

forward checking
constraint satisfaction problem
- variables
- domains
- constraints
  - relation (relace)
  - constraint arity
feasible solution (přípustné řešení)
- complete consistent assignment
constraint modeling
arc consistency
AC-3
look ahead / constraint propagation / maintaining arc consistency
k-consistency
global constraint
heuristics
- variable ordering
  - fail-first principle
- value ordering
  - succeed-first principle

logical reasoning

CNF
- clause
- literal
- formula
DPLL
- pure symbol
- unit clause
SAT optimizations
- component analysis
- variable ordeing
  - degree heuristic
  - activity heuristic
- random restarts
- clever indexing
  - watched literals
- clause learning
knowledge-based agents
- knowledge base
- tell
- ask
- inference
model
entailment
- $X$ entails $Y$
resolution rule
Horn clause
forward chaining
- data-driven reasoning
backward chaining
- goal-driven reasoning

automated planning

fluent
observation model
transition model
- effect axioms
- frame problem
- frame axioms
- successor-state axioms
hybrid planner
propositional planning
- propositional sentence
- SATPlan
- precondition axioms
- action exclusion axioms
- propositional inference
- domain-specific heuristics
situation calculus
- first-order logic
- situations
- states
- possibility axiom
- successor-state axiom
classical planning
- factored representation
- action schema / operator
  - name
  - preconditions
  - effects
- atoms
  - fluents
  - rigid atoms
- closed world assumption
- action
  - relevant action
- regression set
planning problem
- domain model
- initial state
- goal condition
forward/progression planning
backward/regression planning
heuristic
- relaxed problem
  - ignore action preconditions
  - ignore negative effects
other forms of planning
- plan-space
  - open goals
  - casual threats
- hierarchical
  - decompose tasks

probabilistic reasoning

uncertainity
- partial observability
- nondeterminism
logical agent
- belief states
- contingency plans
probabilistic agent
sample space
conditional/posterior probability
product rule
random variables
full joint probability distribution
marginalization
normalization
independence
enumeration
Bayes' rule
casual direction
diagnostic direction
naive Bayes model
chain rule
Bayesian network
- conditional probability distribution
hidden variables
dynamic programming
factors
- product of factors
- summing out a variable
Monte Carlo
- rejection sampling
- likelihood weighting

probabilistic reasoning over time

world state
- time slices / states
  - hidden random variables
  - observable random variables
- discrete time
formal model
- transition model
  - Markov assumption
  - stationary process
- sensor/observation model
  - sensor Markov assumption
basic inference tasks
- filtering
- prediction
  - mixing time
- smoothing
- most likely explanation
  - Viterbi algorithm
hidden Markov model
- matrix implementation
- forward messager propagation
- backward message propagation
dynamic Bayesian network
- unrolling
- exact inference
  - variable elimination
- approximate inference
  - likelihood weighting

decision making

utility
expected utility
maximum expected utility
lottery
preference
normalized utility function
- standard lottery
decision network / influence diagram
- evidence variables
- chance nodes
- decision nodes
- utility nodes
decision-theoretic expert systems
- causal model
- qualitative decision model
- assign probabilities
- assign utilities
- verify and refine
- sensitivity analysis
Markov decision process
- sequence decision problem
- observable stochastic environment
- Markov transition model
- reward
- utility function
  - discount factor
- policy
  - optimal policy
- true utility of a state
- Bellman equation
  - nonlinear
- value iteration
- policy iteration
- partially observable Markov decision process
  - belief states

games and multi-agent systems

simplest games
- deterministic
- turn taking
- two-player
- zero-sum
- games of perfect information
game tree
- minimax
- alpha-beta pruning
- evaluation function
- expected minimax
single-move games
- payoff function
- optimal strategy
- pure strategy
- mixed strategy
- solution
- prisoner's dilemma
  - testify/defect
  - refuse/cooperate
- Nash equilibrium
- Pareto dominated
- Pareto optimal outcome
- two-finger Morra
  - maximin
- repeated prisoner's dilemma
  - tit-for-tat
  - perpetual punishment
mechanism design
- bidder
- private value
- common value
- good mechanism
  - efficient
- dominant strategy
- English auction
  - ascending-bid
- Dutch auction
  - descending-bid
- sealed-bid auction
- sealed-bid second-price auction
  - Vickrey auction
- tragedy of commons
  - Vickrey-Clarke-Groves mechanism
solved game

machine learning

unsupervised learning
reinforcement learning
supervised learning
- hypothesis
- hypothesis space
- consistent hypothesis
- Ockham's razor
- classification
- regression
- decision trees
  - sequence of tests
  - divide-and-conquer strategy
  - entropy
  - information gain
- logic formulation of learning
  - logical classification
  - false negative
  - false positive
  - curent-best-hypothesis search
  - least-commitment search
    - version space
    - G-set / most general boundary
    - S-set / most specific boundary
- linear models
  - univariate linear function
  - multivariate linear function
  - square loss function
  - linear regression
    - analytical approach
    - gradient descent method
      - learning rate
  - linear classification
    - linear separator
    - treshold function
      - perceptron learning rule
      - logistic treshold function
        
        gradient descent
- artificial neural network
  - connected nodes/units
    - inputs
    - activation function
      - perceptron
      - sigmoid perceptron
    - output
  - weighted directional links
- neural network structures
  - feed-forward network
  - recurrent network
- feed-forward network learning
  - gradient descent
  - back-propagation
- parametric model
- non-parametric model
  - nearest neighbor models
  - Minkowski distance
- support-vector machine
  - maximum margin separator
  - kernel function

learning probabilistic models and reinforcement learning

Bayesian learning
- Bayesian networks
- parameter learning
- maximum-likelihood parameter learning
- expectation-maximization algorithm
reinforcement learning
- reward/reinforcement
  - input percept
- passive learning
  - fixed policy
  - direct utility estimation
    - expected reward-to-go
    - running average for each state
  - adaptive dynamic programming
    - transition model
    - rewards
  - temporal-difference learning
    - learning rate
    - model-free method
- active leraning
  - active adaptive dynamic programming agent
    - greedy agent
  - exploration
  - exploitation
  - exploration policies
    - Q-learning
      - temporal-difference learning
      - off-policy
    - SARSA
      - temporal-difference learning
      - on-policy

philosophical and ethical issues

people
- certainity effect
- ambiguity aversion
AI
- weak
- strong
- general
- narrow
Turing test
Gödel's incompleteness theorem
mind-body problem
- dualist theory
- monist theory
thought experiments
- brain in a vat
- brain replacement
- Chinese room
- trolley problem