Machine Learning and ILP for Multi-Agent Systems

Why Learning Agents?

A Brief History

Outline

What is Machine Learning?

Types of Learning

Inductive Learning

Inductive Learning

Inductive Learning Example

Performance Measure

Where’s the knowledge?

Example Language

Hypothesis Language

Learning bias

Inductive Learning

Inductive Learning for Agents

Batch vs Incremental Learning

Batch Learning for Agents

Eager vs. Lazy learning

Active Learning

Black-Box vs. White-Box

Reinforcement Learning

Q Learning

Q Learning

Q Learning

Pros and Cons of RL

Combination of IL and RL

Unsupervised Learning

Learning and Verification

Learning and Verification
[Gordon ’00]

Learning and Verification

Learning in Multi-Agent Systems

Types of Multi-Agent Learning
[Weiss & Dillenbourg 99]

Social Awareness

Levels of Social Awareness [Vidal&Durfee 97]

Social Awareness and Q Learning

Agent models and Q Learning

Agent Models and Q Learning

Q Learning and Communication
[Tan 93]

Role Learning

Q Learning of roles

Q Learning of Roles
[Balch 99]

Distributed Learning

Distributed Data Mining

Bibliography

Bibliography

B R E A K

Machine Learning and ILP for MAS: Part II

Machine Learning and ILP for MAS: Part II

From Machine Learning to Learning Agents

Integrating Machine Learning into the Agent Architecture

Time Constraints on Learning

Doing Eager vs. Lazy Learning
under Time Pressure

 “Clear-cut” vs. Any-time Learning

Time Constraints on Learning in Simulated Environments

Synchronisation ´ Time Constraints

Learning and Recall

Learning and Recall (2)

Learning and Recall (3)

Learning and Recall (4)

Machine Learning and ILP for MAS: Part II

Machine Learning Revisited

Object and Concept Language

Machine Learning Biases

Preference Bias, Search Bias & Version Space

Inductive Logic Programming

LP as ILP Object Language

ILP Object Language Example

LP as ILP Concept Language

Modes in ILP

Modes in ILP

Modes in ILP

Modes in ILP

Types in ILP

ILP Types and Modes: Example

Positive Only Learning

Background Knowledge

Background Knowledge (2)

  Choice of Background Knowledge

ILP Preference Bias

Induction in ILP

Example of Induction

Induction in Progol

Summary of ILP Basics

Learning Pure Logic Programs vs. Decision Lists

Decision List Example

Updating Decision Lists with Exceptions

Updating Decision Lists with Exceptions

Replacing Exceptions with Rules:  Before

Replacing Exceptions with Rules: After

Eager ILP vs. Analogical Prediction

Analogical Prediction Example

Analogical Prediction Example

Timing Analysis of Theories Learned with ILP

Timing Analysis of ILP Theories: Example

Machine Learning and ILP for MAS: Part II

Agent Applications of ILP

Agent Applications of ILP

Agent Applications of ILP

Agent Applications of ILP

The York MA Environment

The York MA Environment

The York MA Environment

Machine Learning and ILP for MAS: Part II

Learning and Natural Selection

Darwinian vs. Lamarckian Evolution

Darwinian vs. Lamarckian Evolution (2)

Learning and Language

Communication and Learning

Communication and Learning (2)

Communication and Learning (3)

Our Current Research

The End