Cognitive Radio Architecture: The Engineering Foundations of Radio XML

Chapter 4. AUTONOMOUS MACHINE LEARNING FOR AACR

This chapter develops AML as the core technology for AACR mass customization, focusing on reinforcement learning in which the iCR independently detects the learning opportunity, shapes the dialog and proactively verifies the acquisition of enhanced skill without burdening the user or violating radio etiquette.

MACHINE LEARNING FRAMEWORK

Machine learning texts typically develop ML strategies, data structures, algorithms, and parameter tuning for relatively simple problems so that the student may understand the ML method. Such simple problems include blocks-world, Rubik's Cube, and Towers of Hanoi [287]. More realistic examples include learning the structures of cells, inducing natural language grammars [93], balancing a rod, and exploring a maze [92]. Examples of ML applications to important problems include medicine, data mining, wireless channel coding [63], and cellular network admissions control [92] among others. Proceedings collect technical papers for emerging topics, such as agent technology in telecommunications [94] and ML in SDR [95].

This chapter applies relevant ML techniques to AACR as a self-contained overview of AML for AACR evolution, with knowledge objects (KOs) and domain heuristics (DHs) developed in subsequent chapters.

The AACR ML Framework

Algorithms that learn may be parametric, defined over continuous domains; or symbolic, defined over discrete domains; or both. Learning to recognize the difference between a speech ...