Work package 4

Task 4.1 – Reasoning about Compute Continuum

This task enriches the Unified Knowledge Layer (UKL) by integrating advanced knowledge representation and reasoning techniques, such as matching, consistency checking, and instance-based learning.

It implements real-time monitoring to analyze observability data and diagnose the state of the SPICE platform or user applications. A reasoning engine will apply multi-objective optimization to meet user constraints and ensure consistency through ontology-based validation.

---

Task 4.2 – AI-Based Methods for Applications and Autonomous Platform Adaptation

This task enhances the UKL with dynamic AI and machine learning algorithms, including sequence-to-sequence learning (RNNs, encoder-decoder models, transformers, Markov models) and reinforcement learning.

Continual, incremental, and lifelong learning approaches will allow the system to adapt to evolving data and address concept drift a catastrophic forgetting. Models will operate at runtime, using active learning for real-time adaptation and performance optimization.

---

Task 4.3 – Platform-Agnostic Perceptive Composer (PPC) for Complex Data Pipelines

This task develops an AI-augmented, semi-automatic pipeline composer that uses XaaS services from the Marketplace (T7.4). It combines lightweight semantics, semantic graphs, ontologies, and machine/deep learning (from the UKL) with historical monitoring data (from T5.2, T6.1, T7.3).

A Visual Pipeline Editor (VPE) will provide an intuitive GUI for pipeline creation. The resulting pipeline definitions will be stored in the Pipeline Catalogue (part of the UKL) and used by the Mapper for mapping and execution.

---

Task 4.4 – User Dashboard

The User Dashboard (UD) enables seamless interaction with the SPICE platform via a user-friendly web portal. Developers can register implementations, add semantic annotations, and contribute to the knowledge base.

It integrates multiple module interfaces to ensure consistency and usability. Users can define optimization criteria, constraints, and non-functional requirements, while administrators manage data access, functions, and security policies.

Secure authentication is ensured through OpenID Connect a OAuth, supporting robust identification for users, processes, and applications.