Page 82 - Artificial Intelligence for the Internet of Everything

68 Artificial Intelligence for the Internet of Everything

In this chapter we address two fundamental issues in IoE. First, we
describe a general framework of adaptive multiagent behavior based
on minimizing a team’s free energy. This framework explains how multiple
autonomous agents can produce team-optimal context-aware behaviors
by performing collaborative perception and control. Second, we present a
mechanism for IoE agents to instantiate adaptive behaviors by intelli-
gently sampling their environment and changing their organization structure. This
structure adaptation modifies the agents’ roles and relations, which encode
and constrain their decision responsibilities and interactions, and is com-
puted in a distributed manner without a central authority. Energy optimi-
zation formally enables locally computed but globally optimal decisions by
using approximate variational inference. The agents make local decision and
communicate by passing belief messages in peer-to-peer manner. By provid-
ing the formal mapping between adaptive decisions, goal-driven actions, and
perception, this model prescribes foundational functional requirements for
developing IoE entities and networks that can efficiently operate in the com-
plex, dynamic, and uncertain environments of the future.

4.2 ENERGY-BASED ADAPTIVE AGENT BEHAVIORS
4.2.1 Free Energy Principle
Recently, Friston proposed a theory, called the free energy principle, that
describes how the agents and biological systems (such as a cell or a brain)
adapt to the uncertain environments by reducing the information-theoretic
quantity known as “variational free energy” (Friston, 2010; Friston, Thorn-
ton, & Clark, 2012). This theory brings Bayesian, information-theoretic,
neuroscientific, and machine-learning approaches into a single formal
framework. The framework prescribes that agents reduce their free energy
in three ways: (1) by changing sensory inputs (control); (2) by changing pre-
dictions of the hidden variables and future sensory inputs (perception); and (3)
by changing the model of the agent, such as its form, representation of envi-
ronment, and structure of relations with other agents (learning and
reorganization).
Variational free energy is defined as a function of sensory outcomes and a
probability density over their (hidden) causes. This function is an upper
bound on surprise, a negative log of the model evidence representing the dif-
ference between an agent’s predictions about its sensory inputs, and the
observations it actually encounters. Since the long-term average of surprise
is entropy, an agent acting to minimize free energy will implicitly place an