Dynamic Bayesian Collective Awareness Models for a Network of Ego-Things

Abstract

A novel approach is proposed for multimodal collective awareness (CA) of multiple networked intelligent agents. Each agent is here considered as an Internet-of-Things (IoT) node equipped with machine learning capabilities; CA aims to provide the network with updated causal knowledge of the state of execution of actions of each node performing a joint task, with particular attention to anomalies that can arise. Datadriven dynamic Bayesian models learned from multisensory data recorded during the normal realization of a joint task (agent network experience) are used for distributed state estimation of agents and detection of abnormalities. A set of switching dynamic Bayesian network (DBN) models collectively learned in a training phase, each related to particular sensorial modality, is used to allow each agent in the network to perform synchronous estimation of possible abnormalities occurring when a new task of the same type is jointly performed. Collective DBN (CDBN) learning is performed by unsupervised clustering of generalized errors (GEs) obtained from a starting generalized model. A growing neural gas (GNG) algorithm is used as a basis to learn the discrete switching variables at the semantic level. Conditional probabilities linking nodes in the CDBN models are estimated using obtained clusters. CDBN models are associated with a Bayesian inference method, namely, distributed Markov jump particle filter (D-MJPF), employed for joint state estimation and abnormality detection. The effects of networking protocols and of communications in the estimation of state and abnormalities are analyzed. Performance is evaluated by using a small network of two autonomous vehicles performing joint navigation tasks in a controlled environment. In the proposed method, first the sharing of observations is considered in ideal condition, and then the effects of a wireless communication channel have been

Graph-Powered Interpretable Machine Learning Models for Abnormality Detection in Ego-Things Network

“…Moreover, the work in [ 14 , 15 ] showed a methodology to develop models to represent self-awareness in agents by considering low dimensional as well as high dimensional sensory data. The initial level of interpretability for the ML models represents collective awareness in agents presented in [ 16 ]. Most of the existing state-of-the-art approaches lack in model’s interpretability to show which features are used to make automated decisions.…”

“…This section describes the steps involved in interpretable ML models’ training and testing phases by considering a simple two-ego-things network. Ego-things are the intelligent agents that can perceive their internal status and external environment and can adapt themselves when they face abnormal situations [ 16 ].…”

“…The first objective of the model is to detect abnormalities at multiple abstraction levels (e.g., continuous and discrete levels), then incrementally learn new models from the generalized errors (GEs) coming from these anomaly data. Generalized errors (GEs) [ 16 ] can be defined as the mismatch between the Bayesian predictions and experimental evidence, computed for the hidden generalized variables of the generative model, i.e., the hidden states and the deviations of states’ together (i.e., generalized states). The higher-order derivatives are limited to first-order, and the GEs are computed through the anomaly detection process.…”

“…The MJPF [ 25 ] is a hybrid filter: a Kalman filter at the continuous level and particle filter at the discrete level to make inferences. A detailed description of the MJPF can be found in [ 16 , 22 ]. However, in this paper, we have used a modified version of it by adding the functionality to detect discrete level global anomaly along with the local anomaly detection capability.…”