1/39
A set of 40 practice flashcards reviewing core concepts, definitions, architectures, and processes in Expert Systems and Machine Learning (IT2206).
What is an expert system?
An intelligent computer program that uses knowledge and inference procedures to solve problems difficult enough to require significant human expertise.
What is factual knowledge?
Knowledge based on facts.
What is heuristic knowledge?
Knowledge based on practice, evaluation, experiences, and the ability to guess.
How is a knowledge base defined in an expert system?
A collection of information in a given field that stores both factual and heuristic knowledge.
What is the primary function of an inference engine?
Acquires insights from information in the knowledge base.
What does the explanation facility in an expert system do?
Lets the user ask how the system reached a conclusion and why a specific fact is needed; it must justify its advice, analysis, or conclusion.
What is the role of the knowledge acquisition facility?
Elicits, collects, analyzes, models, and validates knowledge for knowledge engineering and management projects.
What is the purpose of the external interface in an expert system?
Lets the expert system work with external files using programs written in conventional languages like C and C++; it is the communication link to the external environment.
How is a database defined in an expert system?
A collection of organized information that can easily be accessed, managed, and updated; the inference engine uses it to hold data while working on a problem.
What does the control system in an inference engine do?
Determines the order of testing in the knowledge base rules.
What is the function of the rule interpreter within an inference engine?
Defines the Boolean application rules.
What does the explanation mechanism do in an inference engine?
Justifies the outcome to the user with the reasoning process.
What is forward chaining?
Moves from the facts of the case toward a goal or conclusion by matching rule conditions (IF) with facts in working memory.
What is backward chaining?
Matches an assumed conclusion or goal with the conclusion (THEN) part of a rule; if a rule is found, its premise becomes the new sub-goal.
What role does the user interface play in an expert system?
The part of the expert system that helps users communicate with it and accomplish their goals efficiently, adapting to users' requirements and work practices.
What is considered the major bottleneck in expert system development?
Knowledge acquisition.
What specific current-task data is stored in an expert system's database?
User answers, outside data, intermediate reasoning results, and conclusions reached so far.
What is a rule-based system architecture?
An architecture that uses production rules, or if-then rules, each with a conditional part on the left and a conclusion or action part on the right.
What are non-production systems in expert system architecture?
Systems that use more structured representation schemes instead of rules, such as semantic or associative networks, frames, tree structures, and neural networks.
What is a decision tree?
A structure used when knowledge can be arranged top to bottom; it takes an object's properties as input and outputs a Boolean yes/no decision.
In a decision tree, what do the internal nodes, branches, and terminal nodes represent?
Each internal node tests one property, branches represent possible test values, and terminal nodes represent object identities.
What is a blackboard system architecture?
A special type of knowledge-based system whose different knowledge sources communicate through a common information field.
What is the blackboard in a blackboard system architecture?
A global database for sharing input data, partial solutions, alternatives, and final solutions.
What are knowledge sources in a blackboard system architecture?
Independent modules that contain the knowledge needed for problem solving.
What does the control unit do in a blackboard system architecture?
Determines which knowledge sources to execute for an optimal problem solution.
What are neural networks?
Computing systems modeled on the human brain's mesh-like network of interconnected processing elements called neurons.
How is knowledge represented in neural networks?
By connection patterns and by adjusting connection weights.
What is machine learning?
The process of training a model to make useful predictions from data.
What is a machine learning model?
A representation of the mathematical relationship between elements of data that an ML system uses to make predictions.
What is supervised learning?
Learning from labeled examples.
What is unsupervised learning?
Learning from unlabeled data, commonly by clustering items with common properties.
What is reinforcement learning?
Learning by trial and error, with feedback guiding the system toward the best outcome.
What is regression in supervised learning?
A type of supervised learning in which the model predicts the values of a continuous variable.
What is classification in supervised learning?
A type of supervised learning in which the model predicts the likelihood that an item belongs to a distinct set of categories.
What is the difference between a binary classification model and a multiclass classification model?
A binary classification model outputs a value from a class containing only two values, while a multiclass classification model outputs a value from a class containing more than two values.
What are the five core concepts of supervised learning?
Data, model, training, evaluation, and inference.
What are features and labels in supervised learning?
Features are the values used to predict the label, and the label is the answer or value the model has to predict.
What is loss in machine learning?
The difference between a predicted value and the actual value.
What is the goal of training in machine learning?
To make the best predictions on unseen data.
What is the difference between evaluation and inference in supervised learning?
Evaluation uses a labeled dataset while giving the model only its features to compare predictions with true values, whereas inference uses the model to make predictions on unlabeled examples after evaluation is satisfactory.