Artificial Intelligence Study Notes

• Study notes for Artificial Intelligence (21CSC206T)

Assessment Plan

• Cycle Test-I includes a written test (10 marks), quiz/puzzles (5 marks), and AWS online Course Completion (Machine Learning Foundation) with 10 marks
• Cycle Test-II consists of a written test (10 marks), quiz/puzzles (5 marks), and solving 5 questions on Hackerank (10 marks)
• Hackathon/Group Activity involves a Global Challenge/Hackathons/Ideathons/Makethons /Any AI Technical Competitions including conference presentations/ Samsung Prism (5 marks), and Group Activity (Poster Presentation) (5 marks)
• The total marks for assessment is 60

Introduction to AI

• AI helps in making decisions with reduced human intervention, as seen in automated climate control in cars and self-driving cars
• AI holistically includes learning, searching, and problem solving
• The purpose of AI is to enable machines to solve problems intelligently

Definitions of AI

• AI is defined in multiple ways, categorized in two dimensions: thought process/reasoning and behavior

Acting Humanly: The Turing Test Approach

• Natural language processing enables successful communication in English
• Knowledge representation stores information
• Automated reasoning uses stored information to answer questions and draw conclusions
• Machine learning adapts to new circumstances and detects/extrapolates patterns
• Computer vision allows perceiving objects

Thinking Humanly: The Cognitive Modeling Approach

• Cognitive science is an interdisciplinary field bringing together computer models from AI
• Experimental techniques from psychology are used to construct precise, testable theories of the human mind

Thinking Rationally: The "Laws of Thought" Approach

• This is the study of mental faculties through computational models
• Logic struggles to express uncertainty, needs guidance and intelligent behavior controlled by logic

Acting Rationally: The Rational Agent Approach

• Rational agents aim to achieve the best outcome or the best expected outcome in situations of uncertainty
• An agent is some thing that acts

PEAS (Performance, Environment, Actuator, Sensor)

• Performance measure defines the success of an agent
• Varies based on the agent's percepts
• Environment is the surrounding agent at every instant
• Can include 5 major types: Fully Observable & Partially Observable, Episodic & Sequential, Static & Dynamic, Discrete & Continuous, Deterministic & Stochastic.
• Actuator delivers the output/action to the environment
• Sensors are the receptive parts of an agent; they take in the input

Environment: Deterministic vs. Non-Deterministic

• The environment is deterministic if the next state is determined solely by the current state and the agent's actions
• Inaccessible environments can appear non-deterministic due to the agent's limited perception
• The agent's viewpoint must be known when determining determinism because the agent might have limited perception

Environment: Episodic vs. Non-Episodic

• In episodic environments, the agent's experience is divided into independent episodes of percept sequence and action
• The agent doesn't need to know the effect of its actions

Environment: Static vs. Dynamic

• The environment is dynamic if it changes during an agent's response to a percept sequence
• The environment is static if it stays the same while the agent decides on an action
• The agent doesn't need to compensate for time

Environment: Discrete vs. Continuous

• The environment is discrete if the number of percepts and actions within it is limited and distinct

AI Rational Agent Examples

• Hospital Management System: performance is Patient's health, Admission process, Payment and using symptoms
• Automated Car Drive: performance is comfort, safety, max distance and sensors that includes camera, Odometer, and GPS
• Subject Tutoring: performance is Maximized scores, Improvement in students and using the sensor that include Eyes, Ears, Notebooks
• Part-picking robot: performance is percentage of parts in correct bins and output us Jointed arms and hand, using camera, joint angle sensors

Foundations of AI

• Philosophy: Knowledge Representation, Logic, and AI's possibility
• Math: Search, Analysis of search algorithms, and logic
• Economics: Expert Systems, Decision Theory, and Rational Behavior Principles
• Psychology: Behavioristic insights into AI programs
• Neuroscience: Learning, Neural Nets
• Control Theory and Cybernetics: Information Theory & AI, Entropy, Robotics
• Computer Science & Engineering: Systems for AI

AI Techniques

• AI deals with practical problems, identification/authentication, interdependent/cross-domain issues, and classification
• AI is needed for analysis of large data from multi-domains, characterization/mapping of miscellaneous data, and handling changing scenarios

AI Main Objective

• Capture knowledge based on data and information

AI Task

• Handle different problems, including structured (defined goal state), unstructured (goal state not known), and linear problems (based on dependent variables)

Problem Solving with AI

• Well-structured problems yield the right answer when the appropriate algorithm is applied
• An ill-structured problem don't yield a particular answer
• Includes challenging due to lack of defined steps and criterion to evaluate the outcome

AI Models

• Creative and intelligent tasks modeled like a maze of paths from an initial node to a resultant node (Dunker's 'maze hypothesis')
• Semiotic Models are based on sign process, signification, or communication
• Statistical Models use representation and formalization of relationships through statistical techniques and probabilistic approaches

Data Acquisition/Learning Aspects in AI

• Data Mining and Machine Learning are used in Knowledge Discovery
• Knowledge mining includes extracting meaningful information
• Data mining involves data cleaning, preprocessing, identifying and interpreting patterns, understanding applications, and generalizing target data with consolidated patterns

Machine Learning is making a machine intelligible; based on past experience

• Computational Learning Theory (COLT) uses formal mathematical models to analyze efficiency/complexity in computation, prediction, and feasibility

Neural and Evolutionary computation

• Evolutionary Computation speeds up data mining.
• Neural computing stimulates the neural behavior of humans to enable machine learning
• Artificial Neural Network (ANN) is made for applications like pattern recognition
• Multi-perspective integrated intelligence using knowledge from different perspectives to build intelligent systems

Intelligent Agent and Multi-Agent systems

• Intelligent agents are flexible in acting to get desired outcomes.
• MAS (Multi agent System) involves using more than one intelligent agent to solve complex tasks
• Information collected, it can be continuous or discrete

Problem Solving in AI

• Problem-solving is the process of generating solutions for a given situation
• The problem: defined in a context, with a well-defined objective and has a solution the set of activities
• Problem-solving uses previous and domain knowledge

Types of General purpose and Special purpose problem Solving

• General purpose: comparing situations with goals, selecting actions to reduce the difference
• Special purpose: solving problems that have specific features

Problem Solving Techniques

• Involves, problem definitions, analysis, representation, planning, executing, evaluating, and consolidating gains

Formulating Problems

• A problem is a collection of information that the agent uses to decide what to do for well-defined problems and solutions
• Initial state is the state that the agent knows itself to be
• Goal test determines if something is a goal state

Problem types include:

• Single state (deterministic, observable), multi-state (non-observable)
• Contingency (non-deterministic, partially observable), and exploration (unknown state space)

Measuring Problem-Solving Performance

• Measurement is in whether or not a solution is obtained, the quality of the solution, search cost, and total search cost
• Search cost is associated with the time and memory required to find a solution

To choose an appropriate method for any type Problem

• Is the problem decomposable?
• Can solution steps be ignored or undone?
• Is the universe predictable?
• Is a good solution absolute or relative?
• Is the solution a state or a path?
• What is the role of knowledge?
• Does the task require human-interaction?

Important questions to determining which method is needed

• Is the problem decomposable into sub-problems easy to solve?
• Can solution steps be ignored or undone?
• Is the universe of the problem is predictable?
• Is the problem solution absolute or relative?
• Is the solution a state or a path?
• What is the role of knowledge?
• Does the task require human-interaction?

Role of Knowledge

• Solitary and conversational are types of problems
• Solitary no intermediate communication
• Conversational type is intermediate

Search

• Search is a general algorithm helping finds the path in state space
• The path may lead to the solution or dead end
• Forward search is data-directed (starts from initial state)
• Backward search is goal-directed (starts from target state)

Strategies to explore the states

• Informed search has no guarantee for solution but high probability of getting solution based on heuristic approach
• Uninformed search generates all states while time consuming due to large state space used where error in the algorithm has severe consequences
• Parameters for search evaluation includes completeness (Guaranteed to find a solution in time), space/time complexity, and optimality/admissibility(correctness of the solution)