From a knowledge-level perspectively, the general problem solving capability emerges from an interaction of declarative and procedural knowledge components. In agent-based system, learning is achieved from the interplaying between Memory and Behavior, in which memory is used for storing/retrieving declarative elements, and behavior is used for enhancing memory by executing the actions in procedural components.

Under the MAOS framework, most components and the relations between them are hard-coded so as to reduce the complexity for maintaining MAOS. Moreover, each knowledge component is can be instantiated from a hard-coded JAVA class file by the corresponding TYPE and setting parameters.

Memory Elements

In the mini series, each memory only contains one declarative element, and thus it is reduced to the element it owns. Specifically, M_A contains one EncodedState, called baseState, M_S contains one IGetEachEncodedStateEngine, called referEngine, and M_G contains one EncodedState, called trialState.

Each EncodedState is an instance of maosKernel.memory.EncodedState, which contains a SearchState & its quality, is used (so as not to calculate the quality for a same state at each time).

Each IGetEachEncodedStateEngine is an instance of maosKernel.memory.IGetEachEncodedStateEngine. It contains a set of EncodedStates.

Here SearchState is defined by maosKernel.represent.landscape.space.SearchState. The SearchStates in all the memories (insides baseState, referEngine, and trialState) belong to SPACE_SEARCH, i.e., they are potential solutions themselves. Please refer to the document for Search Spaces for more information.

Behavioral Components

Each behavioral component is a hard-coded file containing certain actions operated on input/output parameters, and certain setting parameters for adjusting operations. All parameters are represented in the format [NAME=VALUE].

All behavioral components are extended from maosKernel.behavior.AbsBehavior. They are located in the dictionaries "maosKernel.behavior", "implement/common/behavior", and "implement/$PROBTYPE/behavior".

Search Components: Single-State Search (Scratch Search, State Repairer, Perturbation Search, Local Search), Recombination Search, Complex Search

Auxiliary Components: State Picker, State Updater

Components for Interaction Center: Topology

Basic Interfaces

There are three flexible interfaces for behavioral components associated with the solving behaviors of agents, i.e., R_INI (initializing), R_G (generating), and R_T (testing), and one interface associated with the coordinating behavior of the Interaction Center (IC) in the environment, i.e., TOPO (interaction protocol / connection topology).

Name	Type	Description
R_INI	maosKernel.behavior.scratch.AbsScratchSearch	For generating an search state from scratch. It only works at initialization stage.
R_G	maosKernel.behavior.generate.AbsMiniGenerator	For generating an search state by using memory elements. It works during the runtime.
R_T	maosKernel.behavior.update.AbsStateUpdater	For updating the generated state into memory. It works during the runtime.
TOPO	maosKernel.behavior.topology.AbsTopology	Maintain the connection topology among the agents. It works during the runtime.