Piglet's Home Page/Computer Science Home
COSC 343 Notes
Artificial Intelligence
SECTION I: REACTIVE MACHINES
- WHAT IS AI?
Turing test, Chinese room.
- STIMULUS RESPONSE AGENTS
- Perception vs Action
eg. Braitenberg vehicle - same vs cross-connections
Production systems: ci -> ai
eg. lift - sensations: button pushed, floor, weight, direction, between doors?
To build up complex behaviour, move backwards
- Complexity
| | simple | complex |
| behavioural | avoid light, green<->red | move through maze |
| algorithmic | keep hand on left wall | traffic lights at Sydney Olympics |
| information |
=> when design system, analyse both the agent and the environment
- Emergent Behaviours
not obvious with individuals
eg. grasshoppers -> swarm of locusts
flocking, traffic jams(-ve effects)->red light in tunnel, behive, nanomachines
- Subsumption Architecture
- hierarchical - behaviours subsume others
- horizontal, layered response, bottom-up approach
- sensations -> distinct perceptions -> actions
- more urgent perceptual events (eg. monster) override less important (eg. eating)
- eg. T-receptor in pigeon eye o->O fkues with 1s to impact
- SINGLE LOGIC UNITS
- Threshold Logic Units (TLU's)
- Credit Assignment Problem
- Delta Rule / Delta Learning
- NEURAL NETWORKS
- Feedforward / Associated Networks
- Training Multilayer Networks
- Momentum
- Structural Parameters
- Bidirectional Networks / Autoassociators
- Preprocessing of Data
- Distributed Representation
- Advantages / Disadvantages of Neural Networks
- TYPES OF LEARNING
- Supervised
- Unsupervised
- Reinforcement
- Training
- GENETIC ALGORITHMS
- MEMORY
- Path Invariant Systems
- Types of Memory
- Iconic
- Feature-based
- Episodic
- Finite State Machines
- Blackboard Systems
- Markov Property
- Updating to Prevent Anomalies
- SIGNAL PROCESSING IN VISION
- Evolution of
- Image to Image Transforms (Image Analysis)
- Removing noise / averaging
- Edge enhancement
- Fourier transforms
- Hough transforms
- Image to Freature Transforms (Scene Analysis)
- Segmentation
- Line processing
- Motion and depth
- Model-based matching
- Binding problem
SECTION II: PLANNING AND SEARCH
- SYMBOLIC AI AND PLANNING AGENTS
- Symbolic/Logical AI
- Sub-symbolic/Analogical/Connectionist AI
- SEARCH IN STATE SPACES - PLANNING
- State Space Graphs
- Uninformed/'Brute Force' Searches
- Depth-first search
- Breadth-first search
- Bounded depth-first search
- Iterative deepening
- Heuristic/Informed Searches
- Evaluation function, f hat
- Greedy search algorithm
- Best-first search
- A* alorithm
- Planning in the Real World
- Sense-plan-act cycle
- Approximate search
- Island-driven search
- Hierarchical search
- Limited horizon search
- Searching off-line
- Learning Heuristic Functions
- Initialize h hat = 0 for all nodes
- Learning h hat in large graphs
- Adversarial Search
- Minimax procedure
- Alpha-beta pruning
SECTION III: KNOWLEDGE REPRESENTATION AND REASONING
- THE PROPOSITIONAL CALCULUS
- Reasoning as Search
- Syntax of Propositional Calculus
- Rules of Inference, R
- Proof, D |- R wn
- Semantics/Meaning -> Interpretation
- Satisfiability and Interpretations
- Entailment, D |= w
- Soundness and Completeness
- Resolution
- Resolution Refutation
- Limitations of Propositional Calculus
- THE PREDICATE CALCULUS
- Syntax of Predicate Calculus
- Semantics -> Interpretation
- Models and Entailment
- Open and Closed wffs
- Unification by Substitution
- Predicate Calculus Resolution (a stronger version)
- Converting wffs to Clause Form
- Tractability -> Horn Clauses
- Limitations of Predicate Calculus
- KNOWLEDGE-BASED SYSTEMS
- Expert Systems
- Commonsense Knowledge
- Taxonomic Knowledge -> Semantic Networks
- NATURAL LANGUAGE PROCESSING
- Utterance as Action
- Phrase-Structure Grammars and Parsing
- Ambiguities
Piglet's Home Page/Computer Science Home
