Activity-Based Credit Assignment Heuristic for Simulation-Based Stochastic Search in a Hierarchical Model Base of Systems

Abstract

Synthesis of systems constitutes a vast class of problems. Although machine learning techniques operate at the functional level, little attention has been paid to system synthesis using a hierarchical model-base. This paper develops an original approach for automatically rating component systems and composing them according to the experimental frames in which they are placed. Components are assigned credit by correlating measures of their participation (activity) in simulation runs with run outcomes. These ratings are employed to bias component selection in subsequent compositions.

Strategy inference during learning via cognitive activity-based credit assignment models

“…In the sequel, learning always refers to a cognitive process, that is the learning of a given task by an individual (or the corresponding model), whereas inference always refers a statistical procedure (eventually combined with an experimental protocol) to guess the learning strategy. The two contributions of this work consist of: (i) a new inference procedure to test different models or hypotheses about how agents learn using only individual observed choice behavior, and (ii) a new version of Activity-Credit Assignment 6 , named Cognitive ACA (CoACA), to model the different possible cognitive strategies used during the learning. In short, we describe strategy-based learning models and an inference procedure - based on cross validation - to select the best model based upon empirical observations.…”

Discrete-Event Simulation Model Generation based on Activity Metrics

“…Activity is a concept and refers to the state transition distribution in the components of a system [15,16]. Activity metrics have been used to speed up simulation in the form of activity-tracking,which focuses computational resources on components based on their activity level.…”

“…Within the context of ongoing efforts presented in [14], the proposed approach is based on the concept of activity [15,16]. This concept is used to define metrics allowing to measure the activity of simulation models.…”

MBSE with/out Simulation: State of the Art and Way Forward

“…A wide variety of mechanisms is available at different levels of abstraction and computational complexity with typical parallels drawn to biologically inspired learning and evolutionary processes including activity-based credit assignment, unsupervised techniques (e.g., clustering, rule mining) and reinforcement learning [42][43][44][45]. These can be based on the premise that new system states are being continuously captured in timely snapshots of data and added to an accumulated repository representing the system knowledge supporting iterative training employing updates in the system behavior.…”

“…Such "design for adaptive sustainment" objectives call for inclusion of models of the environment in which adaptation is occurring as well as of the mechanism mediating the process. A wide variety of such representations is available at different levels of abstraction and computational complexity with typical parallels drawn to biologically inspired learning and evolutionary processes [43][44][45]. One possibility that seems especially apt here (and is rarely considered) rests on the analogy to the carrying capacity of an ecosystem for a member species.…”