A Vector Symbolic Approach for Cognitive Services and Decentralized Workflows

Abstract The proliferation of smart devices and sensors known as the Internet of Things (IoT), along with the transformation of mobile phones into powerful handheld computers as well as the continuing advancement in high-speed communication technologies, introduces new possibilities for collaborative distributed computing and collaborative workflows along with a new set of problems to be solved. However, traditional service-based applications, in fixed networks, are typically constructed and managed centrally and assume stable service endpoints and adequate network connectivity. Constructing and maintaining such applications in dynamic heterogeneous wireless networked environments, where limited bandwidth and transient connectivity are commonplace, presents significant challenges and makes centralized application construction and management impossible. The key objective for this thesis can be summarised as follows: a means is required to discover and orchestrate sequences of micro-services, i.e., workflows, on-demand, using currently available distributed resources (compute devices, functional services, data and sensors) in spite of a poor quality (fragmented, low bandwidth) network infrastructure and without central control. It is desirable to be able to compose such workflows on-the-fly in order to fulfil an ‘intent’. The research undertaken investigates how service definition, service matching and decentralised service composition and orchestration can be achieved without centralised control using an approach based on a Binary Spatter Code Vector Symbolic Architecture and shows that the approach offers significant advantages in environments where communication networks are unreliable. The outcomes demonstrate a new cognitive workflow model that uses one-to-many communications to enable intelligent cooperation between self-describing service entities that can self-organise to complete a workflow task. Workflow orchestration overhead was minimised using two innovations, a local arbitration mechanism that uses a delayed response mechanism to suppress responses that are not an ideal match and the holographic nature of VSA descriptions enables messages to be truncated without loss of meaning. A new hierarchical VSA encoding scheme was created that is scaleable to any number of vector embeddings including workflow steps. The encoding can also facilitate learning since it provides unique contexts for each step in a workflow. The encoding also enables service pre-provisioning because individual workflow steps can be decoded easily by any service receiving a multicast workflow vector. This thesis brings the state-of-the-art closer to the ability to discover distributed services on-the-fly to fulfil an intent and without the need for centralised management or the imperative definition of all service steps, including locations. The use of a mathematically deterministic distributed vector representation in the form of BSC vectors for both service objects and workflows enables a common language for all elements required to discover and execute workflows in decentralised transient environments and opens up the possibilities of employing learning algorithms that can advance the state-of-the-art in distributed workflows towards a true cognitive distributed network architecture.
Authors
  • Chris Simpkin (Cardiff)
Date Jan-2021