Page Comparison

Adaptive Multi-Agent System is based on the concept of "agents". Its global function is to converge towards a state where user-defined criteria are satisfied.

In Pricefx, AMAS is called Multi-Agent Optimization Engine (OE). 

An autonomous program that indefinitely loops its lifecycle of perception-decision-action. “Autonomous” means the agent decides itself what to do, when to do it, and how to do it.

Constraints Parameters

Constraints are threshold and target criteria. Their parameters are namely: 

• Acceptable delta – Acceptable margin of error.

• Precision – Expected precision of the target.

More details in Criteria Description

An agent which takes variables as input to compute a value that it assigns to the output variable. This is one of the concepts of the meta-model.

Associate dimensions (e.g. product, client) to categories (e.g. product X, client Z) to situate variables in the problem's space. This is one of the concepts of the meta-model.

Criterion 

It is an agent which expresses a criticality level by reflecting their satisfaction and their importance, e.g. targets, thresholds, or mini/maximization objectives. Criteria are applied to a variable. This term is one of the concepts of the meta-model.

They are what defines levers and criteria. For instance, if your levers are product prices and your criteria are margin targets per product family and customer, then your dimensions are product, product family, and customer.

The smallest group of the most influential Value Finders whose cumulated influence is greater than half of the total influence received by the Criterion Agent.

Simple computation graph patterns to test the system and identify some Non-Cooperative Situations.

They include SISO – Single input single output, SIMO – Single input multiple outputs, MISO – Multiple input single output, MIMO – Multiple input multiple outputs, Asynchronous feedback, Phase shift, Phase interference, Cycle.

Multi-Agent System which performs multi-dimensional multi-criteria optimization.

The AMAS approach states that agents must be cooperative: they should seek to help each other, adopt a useful behavior, and communicate useful information in a non-ambiguous form. This is the way to ensure that local behaviors lead to an adequate global function. Some situations may prevent agents to cooperate. They are called Non-Cooperative Situations (NCSs) and are of different types depending on what is going wrong at what part of the lifecycle (e.g. conflict, uselessness, ambiguity, etc.). An agent should be able to either detect and solve NCSs or anticipate and avoid them.

This documentation mentions the following types of NCSs: Computation Agent Conflict (Phase Shift), Value Finder Conflict (El Farol Bar Problem), Value Finder Unproductiveness (No Neighbor), Value Finder Incompetence (Asynchronous Feedback), Criterion Agent Conflict (Gradient Loss).

Problem Modeling

Located within a space, it defines the perimeter in which the following variables and criteria exist.

Defines the dimensions in which the following variables and criteria exist (i.e. a plate in the graphical norm).

Defines when Multi-Agent Optimization Engine should stop. The types of conditions are

the maximum number of steps, the minimum dissatisfaction level, and the minimum entropy level. 

Settings for certain types of Value Finders. The settings include initial value, initial/minimal/maximal amplitudes, and increase/decrease coefficients.